Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+.github/preview.png filter=lfs diff=lfs merge=lfs -text

.github/FUNDING.yml

@@ -0,0 +1 @@
+custom: [ buymeacoffee.com/facefusion, ko-fi.com/facefusion ]

.github/workflows/ci.yml

@@ -0,0 +1,58 @@
+name: ci
+
+on: [ push, pull_request ]
+
+jobs:
+ lint:
+  runs-on: ubuntu-latest
+  steps:
+  - name: Checkout
+    uses: actions/checkout@v4
+  - name: Set up Python 3.12
+    uses: actions/setup-python@v5
+    with:
+     python-version: '3.12'
+  - run: pip install flake8
+  - run: pip install flake8-import-order
+  - run: pip install mypy
+  - run: flake8 facefusion.py install.py
+  - run: flake8 facefusion tests
+  - run: mypy facefusion.py install.py
+  - run: mypy facefusion tests
+ test:
+  strategy:
+   matrix:
+    os: [ macos-latest, ubuntu-latest, windows-latest ]
+  runs-on: ${{ matrix.os }}
+  steps:
+  - name: Checkout
+    uses: actions/checkout@v4
+  - name: Set up FFmpeg
+    uses: AnimMouse/setup-ffmpeg@v1
+  - name: Set up Python 3.12
+    uses: actions/setup-python@v5
+    with:
+     python-version: '3.12'
+  - run: python install.py --onnxruntime default --skip-conda
+  - run: pip install pytest
+  - run: pytest
+ report:
+  needs: test
+  runs-on: ubuntu-latest
+  steps:
+  - name: Checkout
+    uses: actions/checkout@v4
+  - name: Set up FFmpeg
+    uses: FedericoCarboni/setup-ffmpeg@v3
+  - name: Set up Python 3.12
+    uses: actions/setup-python@v5
+    with:
+     python-version: '3.12'
+  - run: python install.py --onnxruntime default --skip-conda
+  - run: pip install coveralls
+  - run: pip install pytest
+  - run: pip install pytest-cov
+  - run: pytest tests --cov facefusion
+  - run: coveralls --service github
+    env:
+     GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.gitignore

@@ -0,0 +1,6 @@
+__pycache__
+.assets
+.caches
+.jobs
+.idea
+.vscode

facefusion/processors/choices.py

@@ -2,7 +2,169 @@ from typing import List, Sequence
 
 from facefusion.common_helper import create_float_range, create_int_range
 from facefusion.filesystem import get_file_name, resolve_file_paths, resolve_relative_path
-from facefusion.processors.types import FaceEnhancerModel, FaceSwapperModel, FaceSwapperSet, FrameColorizerModel, FrameEnhancerModel #DeepSwapperModel, AgeModifierModel, FaceDebuggerItem, ExpressionRestorerModel, LipSyncerModel, FaceEditorModel
+from facefusion.processors.types import AgeModifierModel, DeepSwapperModel, ExpressionRestorerModel, FaceDebuggerItem, FaceEditorModel, FaceEnhancerModel, FaceSwapperModel, FaceSwapperSet, FrameColorizerModel, FrameEnhancerModel, LipSyncerModel
+
+age_modifier_models : List[AgeModifierModel] = [ 'styleganex_age' ]
+deep_swapper_models : List[DeepSwapperModel] =\
+[
+	'druuzil/adam_levine_320',
+	'druuzil/adrianne_palicki_384',
+	'druuzil/agnetha_falskog_224',
+	'druuzil/alan_ritchson_320',
+	'druuzil/alicia_vikander_320',
+	'druuzil/amber_midthunder_320',
+	'druuzil/andras_arato_384',
+	'druuzil/andrew_tate_320',
+	'druuzil/angelina_jolie_384',
+	'druuzil/anne_hathaway_320',
+	'druuzil/anya_chalotra_320',
+	'druuzil/arnold_schwarzenegger_320',
+	'druuzil/benjamin_affleck_320',
+	'druuzil/benjamin_stiller_384',
+	'druuzil/bradley_pitt_224',
+	'druuzil/brie_larson_384',
+	'druuzil/bruce_campbell_384',
+	'druuzil/bryan_cranston_320',
+	'druuzil/catherine_blanchett_352',
+	'druuzil/christian_bale_320',
+	'druuzil/christopher_hemsworth_320',
+	'druuzil/christoph_waltz_384',
+	'druuzil/cillian_murphy_320',
+	'druuzil/cobie_smulders_256',
+	'druuzil/dwayne_johnson_384',
+	'druuzil/edward_norton_320',
+	'druuzil/elisabeth_shue_320',
+	'druuzil/elizabeth_olsen_384',
+	'druuzil/elon_musk_320',
+	'druuzil/emily_blunt_320',
+	'druuzil/emma_stone_384',
+	'druuzil/emma_watson_320',
+	'druuzil/erin_moriarty_384',
+	'druuzil/eva_green_320',
+	'druuzil/ewan_mcgregor_320',
+	'druuzil/florence_pugh_320',
+	'druuzil/freya_allan_320',
+	'druuzil/gary_cole_224',
+	'druuzil/gigi_hadid_224',
+	'druuzil/harrison_ford_384',
+	'druuzil/hayden_christensen_320',
+	'druuzil/heath_ledger_320',
+	'druuzil/henry_cavill_448',
+	'druuzil/hugh_jackman_384',
+	'druuzil/idris_elba_320',
+	'druuzil/jack_nicholson_320',
+	'druuzil/james_carrey_384',
+	'druuzil/james_mcavoy_320',
+	'druuzil/james_varney_320',
+	'druuzil/jason_momoa_320',
+	'druuzil/jason_statham_320',
+	'druuzil/jennifer_connelly_384',
+	'druuzil/jimmy_donaldson_320',
+	'druuzil/jordan_peterson_384',
+	'druuzil/karl_urban_224',
+	'druuzil/kate_beckinsale_384',
+	'druuzil/laurence_fishburne_384',
+	'druuzil/lili_reinhart_320',
+	'druuzil/luke_evans_384',
+	'druuzil/mads_mikkelsen_384',
+	'druuzil/mary_winstead_320',
+	'druuzil/margaret_qualley_384',
+	'druuzil/melina_juergens_320',
+	'druuzil/michael_fassbender_320',
+	'druuzil/michael_fox_320',
+	'druuzil/millie_bobby_brown_320',
+	'druuzil/morgan_freeman_320',
+	'druuzil/patrick_stewart_224',
+	'druuzil/rachel_weisz_384',
+	'druuzil/rebecca_ferguson_320',
+	'druuzil/scarlett_johansson_320',
+	'druuzil/shannen_doherty_384',
+	'druuzil/seth_macfarlane_384',
+	'druuzil/thomas_cruise_320',
+	'druuzil/thomas_hanks_384',
+	'druuzil/william_murray_384',
+	'druuzil/zoe_saldana_384',
+	'edel/emma_roberts_224',
+	'edel/ivanka_trump_224',
+	'edel/lize_dzjabrailova_224',
+	'edel/sidney_sweeney_224',
+	'edel/winona_ryder_224',
+	'iperov/alexandra_daddario_224',
+	'iperov/alexei_navalny_224',
+	'iperov/amber_heard_224',
+	'iperov/dilraba_dilmurat_224',
+	'iperov/elon_musk_224',
+	'iperov/emilia_clarke_224',
+	'iperov/emma_watson_224',
+	'iperov/erin_moriarty_224',
+	'iperov/jackie_chan_224',
+	'iperov/james_carrey_224',
+	'iperov/jason_statham_320',
+	'iperov/keanu_reeves_320',
+	'iperov/margot_robbie_224',
+	'iperov/natalie_dormer_224',
+	'iperov/nicolas_coppola_224',
+	'iperov/robert_downey_224',
+	'iperov/rowan_atkinson_224',
+	'iperov/ryan_reynolds_224',
+	'iperov/scarlett_johansson_224',
+	'iperov/sylvester_stallone_224',
+	'iperov/thomas_cruise_224',
+	'iperov/thomas_holland_224',
+	'iperov/vin_diesel_224',
+	'iperov/vladimir_putin_224',
+	'jen/angelica_trae_288',
+	'jen/ella_freya_224',
+	'jen/emma_myers_320',
+	'jen/evie_pickerill_224',
+	'jen/kang_hyewon_320',
+	'jen/maddie_mead_224',
+	'jen/nicole_turnbull_288',
+	'mats/alica_schmidt_320',
+	'mats/ashley_alexiss_224',
+	'mats/billie_eilish_224',
+	'mats/brie_larson_224',
+	'mats/cara_delevingne_224',
+	'mats/carolin_kebekus_224',
+	'mats/chelsea_clinton_224',
+	'mats/claire_boucher_224',
+	'mats/corinna_kopf_224',
+	'mats/florence_pugh_224',
+	'mats/hillary_clinton_224',
+	'mats/jenna_fischer_224',
+	'mats/kim_jisoo_320',
+	'mats/mica_suarez_320',
+	'mats/shailene_woodley_224',
+	'mats/shraddha_kapoor_320',
+	'mats/yu_jimin_352',
+	'rumateus/alison_brie_224',
+	'rumateus/amber_heard_224',
+	'rumateus/angelina_jolie_224',
+	'rumateus/aubrey_plaza_224',
+	'rumateus/bridget_regan_224',
+	'rumateus/cobie_smulders_224',
+	'rumateus/deborah_woll_224',
+	'rumateus/dua_lipa_224',
+	'rumateus/emma_stone_224',
+	'rumateus/hailee_steinfeld_224',
+	'rumateus/hilary_duff_224',
+	'rumateus/jessica_alba_224',
+	'rumateus/jessica_biel_224',
+	'rumateus/john_cena_224',
+	'rumateus/kim_kardashian_224',
+	'rumateus/kristen_bell_224',
+	'rumateus/lucy_liu_224',
+	'rumateus/margot_robbie_224',
+	'rumateus/megan_fox_224',
+	'rumateus/meghan_markle_224',
+	'rumateus/millie_bobby_brown_224',
+	'rumateus/natalie_portman_224',
+	'rumateus/nicki_minaj_224',
+	'rumateus/olivia_wilde_224',
+	'rumateus/shay_mitchell_224',
+	'rumateus/sophie_turner_224',
+	'rumateus/taylor_swift_224'
+]
 
 custom_model_file_paths = resolve_file_paths(resolve_relative_path('../.assets/models/custom'))
 
@@ -12,9 +174,9 @@ if custom_model_file_paths:
 		model_id = '/'.join([ 'custom', get_file_name(model_file_path) ])
 		deep_swapper_models.append(model_id)
 
-#expression_restorer_models : List[ExpressionRestorerModel] = [ 'live_portrait' ]
-#face_debugger_items : List[FaceDebuggerItem] = [ 'bounding-box', 'face-landmark-5', 'face-landmark-5/68', 'face-landmark-68', 'face-landmark-68/5', 'face-mask', 'face-detector-score', 'face-landmarker-score', 'age', 'gender', 'race' ]
-#face_editor_models : List[FaceEditorModel] = [ 'live_portrait' ]
+expression_restorer_models : List[ExpressionRestorerModel] = [ 'live_portrait' ]
+face_debugger_items : List[FaceDebuggerItem] = [ 'bounding-box', 'face-landmark-5', 'face-landmark-5/68', 'face-landmark-68', 'face-landmark-68/5', 'face-mask', 'face-detector-score', 'face-landmarker-score', 'age', 'gender', 'race' ]
+face_editor_models : List[FaceEditorModel] = [ 'live_portrait' ]
 face_enhancer_models : List[FaceEnhancerModel] = [ 'codeformer', 'gfpgan_1.2', 'gfpgan_1.3', 'gfpgan_1.4', 'gpen_bfr_256', 'gpen_bfr_512', 'gpen_bfr_1024', 'gpen_bfr_2048', 'restoreformer_plus_plus' ]
 face_swapper_set : FaceSwapperSet =\
 {
@@ -36,11 +198,11 @@ face_swapper_models : List[FaceSwapperModel] = list(face_swapper_set.keys())
 frame_colorizer_models : List[FrameColorizerModel] = [ 'ddcolor', 'ddcolor_artistic', 'deoldify', 'deoldify_artistic', 'deoldify_stable' ]
 frame_colorizer_sizes : List[str] = [ '192x192', '256x256', '384x384', '512x512' ]
 frame_enhancer_models : List[FrameEnhancerModel] = [ 'clear_reality_x4', 'lsdir_x4', 'nomos8k_sc_x4', 'real_esrgan_x2', 'real_esrgan_x2_fp16', 'real_esrgan_x4', 'real_esrgan_x4_fp16', 'real_esrgan_x8', 'real_esrgan_x8_fp16', 'real_hatgan_x4', 'real_web_photo_x4', 'realistic_rescaler_x4', 'remacri_x4', 'siax_x4', 'span_kendata_x4', 'swin2_sr_x4', 'ultra_sharp_x4', 'ultra_sharp_2_x4' ]
-#lip_syncer_models : List[LipSyncerModel] = [ 'edtalk_256', 'wav2lip_96', 'wav2lip_gan_96' ]
+lip_syncer_models : List[LipSyncerModel] = [ 'edtalk_256', 'wav2lip_96', 'wav2lip_gan_96' ]
 
-#age_modifier_direction_range : Sequence[int] = create_int_range(-100, 100, 1)
-#deep_swapper_morph_range : Sequence[int] = create_int_range(0, 100, 1)
-#expression_restorer_factor_range : Sequence[int] = create_int_range(0, 100, 1)
+age_modifier_direction_range : Sequence[int] = create_int_range(-100, 100, 1)
+deep_swapper_morph_range : Sequence[int] = create_int_range(0, 100, 1)
+expression_restorer_factor_range : Sequence[int] = create_int_range(0, 100, 1)
 face_editor_eyebrow_direction_range : Sequence[float] = create_float_range(-1.0, 1.0, 0.05)
 face_editor_eye_gaze_horizontal_range : Sequence[float] = create_float_range(-1.0, 1.0, 0.05)
 face_editor_eye_gaze_vertical_range : Sequence[float] = create_float_range(-1.0, 1.0, 0.05)
@@ -59,4 +221,4 @@ face_enhancer_blend_range : Sequence[int] = create_int_range(0, 100, 1)
 face_enhancer_weight_range : Sequence[float] = create_float_range(0.0, 1.0, 0.05)
 frame_colorizer_blend_range : Sequence[int] = create_int_range(0, 100, 1)
 frame_enhancer_blend_range : Sequence[int] = create_int_range(0, 100, 1)
-#lip_syncer_weight_range : Sequence[float] = create_float_range(0.0, 1.0, 0.05)
+lip_syncer_weight_range : Sequence[float] = create_float_range(0.0, 1.0, 0.05)

facefusion/processors/modules/age_modifier.py

@@ -0,0 +1,254 @@
+from argparse import ArgumentParser
+from functools import lru_cache
+from typing import List
+
+import cv2
+import numpy
+
+import facefusion.choices
+import facefusion.jobs.job_manager
+import facefusion.jobs.job_store
+import facefusion.processors.core as processors
+from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, inference_manager, logger, process_manager, state_manager, video_manager, wording
+from facefusion.common_helper import create_int_metavar
+from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
+from facefusion.execution import has_execution_provider
+from facefusion.face_analyser import get_many_faces, get_one_face
+from facefusion.face_helper import merge_matrix, paste_back, scale_face_landmark_5, warp_face_by_face_landmark_5
+from facefusion.face_masker import create_box_mask, create_occlusion_mask
+from facefusion.face_selector import find_similar_faces, sort_and_filter_faces
+from facefusion.face_store import get_reference_faces
+from facefusion.filesystem import in_directory, is_image, is_video, resolve_relative_path, same_file_extension
+from facefusion.processors import choices as processors_choices
+from facefusion.processors.types import AgeModifierDirection, AgeModifierInputs
+from facefusion.program_helper import find_argument_group
+from facefusion.thread_helper import thread_semaphore
+from facefusion.types import ApplyStateItem, Args, DownloadScope, Face, InferencePool, ModelOptions, ModelSet, ProcessMode, QueuePayload, UpdateProgress, VisionFrame
+from facefusion.vision import match_frame_color, read_image, read_static_image, write_image
+
+
+@lru_cache(maxsize = None)
+def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
+	return\
+	{
+		'styleganex_age':
+		{
+			'hashes':
+			{
+				'age_modifier':
+				{
+					'url': resolve_download_url('models-3.1.0', 'styleganex_age.hash'),
+					'path': resolve_relative_path('../.assets/models/styleganex_age.hash')
+				}
+			},
+			'sources':
+			{
+				'age_modifier':
+				{
+					'url': resolve_download_url('models-3.1.0', 'styleganex_age.onnx'),
+					'path': resolve_relative_path('../.assets/models/styleganex_age.onnx')
+				}
+			},
+			'templates':
+			{
+				'target': 'ffhq_512',
+				'target_with_background': 'styleganex_384'
+			},
+			'sizes':
+			{
+				'target': (256, 256),
+				'target_with_background': (384, 384)
+			}
+		}
+	}
+
+
+def get_inference_pool() -> InferencePool:
+	model_names = [ state_manager.get_item('age_modifier_model') ]
+	model_source_set = get_model_options().get('sources')
+
+	return inference_manager.get_inference_pool(__name__, model_names, model_source_set)
+
+
+def clear_inference_pool() -> None:
+	model_names = [ state_manager.get_item('age_modifier_model') ]
+	inference_manager.clear_inference_pool(__name__, model_names)
+
+
+def get_model_options() -> ModelOptions:
+	model_name = state_manager.get_item('age_modifier_model')
+	return create_static_model_set('full').get(model_name)
+
+
+def register_args(program : ArgumentParser) -> None:
+	group_processors = find_argument_group(program, 'processors')
+	if group_processors:
+		group_processors.add_argument('--age-modifier-model', help = wording.get('help.age_modifier_model'), default = config.get_str_value('processors', 'age_modifier_model', 'styleganex_age'), choices = processors_choices.age_modifier_models)
+		group_processors.add_argument('--age-modifier-direction', help = wording.get('help.age_modifier_direction'), type = int, default = config.get_int_value('processors', 'age_modifier_direction', '0'), choices = processors_choices.age_modifier_direction_range, metavar = create_int_metavar(processors_choices.age_modifier_direction_range))
+		facefusion.jobs.job_store.register_step_keys([ 'age_modifier_model', 'age_modifier_direction' ])
+
+
+def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
+	apply_state_item('age_modifier_model', args.get('age_modifier_model'))
+	apply_state_item('age_modifier_direction', args.get('age_modifier_direction'))
+
+
+def pre_check() -> bool:
+	model_hash_set = get_model_options().get('hashes')
+	model_source_set = get_model_options().get('sources')
+
+	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)
+
+
+def pre_process(mode : ProcessMode) -> bool:
+	if mode in [ 'output', 'preview' ] and not is_image(state_manager.get_item('target_path')) and not is_video(state_manager.get_item('target_path')):
+		logger.error(wording.get('choose_image_or_video_target') + wording.get('exclamation_mark'), __name__)
+		return False
+	if mode == 'output' and not in_directory(state_manager.get_item('output_path')):
+		logger.error(wording.get('specify_image_or_video_output') + wording.get('exclamation_mark'), __name__)
+		return False
+	if mode == 'output' and not same_file_extension(state_manager.get_item('target_path'), state_manager.get_item('output_path')):
+		logger.error(wording.get('match_target_and_output_extension') + wording.get('exclamation_mark'), __name__)
+		return False
+	return True
+
+
+def post_process() -> None:
+	read_static_image.cache_clear()
+	video_manager.clear_video_pool()
+	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
+		clear_inference_pool()
+	if state_manager.get_item('video_memory_strategy') == 'strict':
+		content_analyser.clear_inference_pool()
+		face_classifier.clear_inference_pool()
+		face_detector.clear_inference_pool()
+		face_landmarker.clear_inference_pool()
+		face_masker.clear_inference_pool()
+		face_recognizer.clear_inference_pool()
+
+
+def modify_age(target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
+	model_templates = get_model_options().get('templates')
+	model_sizes = get_model_options().get('sizes')
+	face_landmark_5 = target_face.landmark_set.get('5/68').copy()
+	crop_vision_frame, affine_matrix = warp_face_by_face_landmark_5(temp_vision_frame, face_landmark_5, model_templates.get('target'), model_sizes.get('target'))
+	extend_face_landmark_5 = scale_face_landmark_5(face_landmark_5, 0.875)
+	extend_vision_frame, extend_affine_matrix = warp_face_by_face_landmark_5(temp_vision_frame, extend_face_landmark_5, model_templates.get('target_with_background'), model_sizes.get('target_with_background'))
+	extend_vision_frame_raw = extend_vision_frame.copy()
+	box_mask = create_box_mask(extend_vision_frame, state_manager.get_item('face_mask_blur'), (0, 0, 0, 0))
+	crop_masks =\
+	[
+		box_mask
+	]
+
+	if 'occlusion' in state_manager.get_item('face_mask_types'):
+		occlusion_mask = create_occlusion_mask(crop_vision_frame)
+		combined_matrix = merge_matrix([ extend_affine_matrix, cv2.invertAffineTransform(affine_matrix) ])
+		occlusion_mask = cv2.warpAffine(occlusion_mask, combined_matrix, model_sizes.get('target_with_background'))
+		crop_masks.append(occlusion_mask)
+
+	crop_vision_frame = prepare_vision_frame(crop_vision_frame)
+	extend_vision_frame = prepare_vision_frame(extend_vision_frame)
+	age_modifier_direction = numpy.array(numpy.interp(state_manager.get_item('age_modifier_direction'), [ -100, 100 ], [ 2.5, -2.5 ])).astype(numpy.float32)
+	extend_vision_frame = forward(crop_vision_frame, extend_vision_frame, age_modifier_direction)
+	extend_vision_frame = normalize_extend_frame(extend_vision_frame)
+	extend_vision_frame = match_frame_color(extend_vision_frame_raw, extend_vision_frame)
+	extend_affine_matrix *= (model_sizes.get('target')[0] * 4) / model_sizes.get('target_with_background')[0]
+	crop_mask = numpy.minimum.reduce(crop_masks).clip(0, 1)
+	crop_mask = cv2.resize(crop_mask, (model_sizes.get('target')[0] * 4, model_sizes.get('target')[1] * 4))
+	paste_vision_frame = paste_back(temp_vision_frame, extend_vision_frame, crop_mask, extend_affine_matrix)
+	return paste_vision_frame
+
+
+def forward(crop_vision_frame : VisionFrame, extend_vision_frame : VisionFrame, age_modifier_direction : AgeModifierDirection) -> VisionFrame:
+	age_modifier = get_inference_pool().get('age_modifier')
+	age_modifier_inputs = {}
+
+	if has_execution_provider('coreml'):
+		age_modifier.set_providers([ facefusion.choices.execution_provider_set.get('cpu') ])
+
+	for age_modifier_input in age_modifier.get_inputs():
+		if age_modifier_input.name == 'target':
+			age_modifier_inputs[age_modifier_input.name] = crop_vision_frame
+		if age_modifier_input.name == 'target_with_background':
+			age_modifier_inputs[age_modifier_input.name] = extend_vision_frame
+		if age_modifier_input.name == 'direction':
+			age_modifier_inputs[age_modifier_input.name] = age_modifier_direction
+
+	with thread_semaphore():
+		crop_vision_frame = age_modifier.run(None, age_modifier_inputs)[0][0]
+
+	return crop_vision_frame
+
+
+def prepare_vision_frame(vision_frame : VisionFrame) -> VisionFrame:
+	vision_frame = vision_frame[:, :, ::-1] / 255.0
+	vision_frame = (vision_frame - 0.5) / 0.5
+	vision_frame = numpy.expand_dims(vision_frame.transpose(2, 0, 1), axis = 0).astype(numpy.float32)
+	return vision_frame
+
+
+def normalize_extend_frame(extend_vision_frame : VisionFrame) -> VisionFrame:
+	model_sizes = get_model_options().get('sizes')
+	extend_vision_frame = numpy.clip(extend_vision_frame, -1, 1)
+	extend_vision_frame = (extend_vision_frame + 1) / 2
+	extend_vision_frame = extend_vision_frame.transpose(1, 2, 0).clip(0, 255)
+	extend_vision_frame = (extend_vision_frame * 255.0)
+	extend_vision_frame = extend_vision_frame.astype(numpy.uint8)[:, :, ::-1]
+	extend_vision_frame = cv2.resize(extend_vision_frame, (model_sizes.get('target')[0] * 4, model_sizes.get('target')[1] * 4), interpolation = cv2.INTER_AREA)
+	return extend_vision_frame
+
+
+def get_reference_frame(source_face : Face, target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
+	return modify_age(target_face, temp_vision_frame)
+
+
+def process_frame(inputs : AgeModifierInputs) -> VisionFrame:
+	reference_faces = inputs.get('reference_faces')
+	target_vision_frame = inputs.get('target_vision_frame')
+	many_faces = sort_and_filter_faces(get_many_faces([ target_vision_frame ]))
+
+	if state_manager.get_item('face_selector_mode') == 'many':
+		if many_faces:
+			for target_face in many_faces:
+				target_vision_frame = modify_age(target_face, target_vision_frame)
+	if state_manager.get_item('face_selector_mode') == 'one':
+		target_face = get_one_face(many_faces)
+		if target_face:
+			target_vision_frame = modify_age(target_face, target_vision_frame)
+	if state_manager.get_item('face_selector_mode') == 'reference':
+		similar_faces = find_similar_faces(many_faces, reference_faces, state_manager.get_item('reference_face_distance'))
+		if similar_faces:
+			for similar_face in similar_faces:
+				target_vision_frame = modify_age(similar_face, target_vision_frame)
+	return target_vision_frame
+
+
+def process_frames(source_path : List[str], queue_payloads : List[QueuePayload], update_progress : UpdateProgress) -> None:
+	reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
+
+	for queue_payload in process_manager.manage(queue_payloads):
+		target_vision_path = queue_payload['frame_path']
+		target_vision_frame = read_image(target_vision_path)
+		output_vision_frame = process_frame(
+		{
+			'reference_faces': reference_faces,
+			'target_vision_frame': target_vision_frame
+		})
+		write_image(target_vision_path, output_vision_frame)
+		update_progress(1)
+
+
+def process_image(source_path : str, target_path : str, output_path : str) -> None:
+	reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
+	target_vision_frame = read_static_image(target_path)
+	output_vision_frame = process_frame(
+	{
+		'reference_faces': reference_faces,
+		'target_vision_frame': target_vision_frame
+	})
+	write_image(output_path, output_vision_frame)
+
+
+def process_video(source_paths : List[str], temp_frame_paths : List[str]) -> None:
+	processors.multi_process_frames(None, temp_frame_paths, process_frames)

facefusion/processors/modules/deep_swapper.py

@@ -0,0 +1,464 @@
+from argparse import ArgumentParser
+from functools import lru_cache
+from typing import List, Tuple
+
+import cv2
+import numpy
+from cv2.typing import Size
+
+import facefusion.jobs.job_manager
+import facefusion.jobs.job_store
+import facefusion.processors.core as processors
+from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, inference_manager, logger, process_manager, state_manager, video_manager, wording
+from facefusion.common_helper import create_int_metavar
+from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url_by_provider
+from facefusion.face_analyser import get_many_faces, get_one_face
+from facefusion.face_helper import paste_back, warp_face_by_face_landmark_5
+from facefusion.face_masker import create_area_mask, create_box_mask, create_occlusion_mask, create_region_mask
+from facefusion.face_selector import find_similar_faces, sort_and_filter_faces
+from facefusion.face_store import get_reference_faces
+from facefusion.filesystem import get_file_name, in_directory, is_image, is_video, resolve_file_paths, resolve_relative_path, same_file_extension
+from facefusion.processors import choices as processors_choices
+from facefusion.processors.types import DeepSwapperInputs, DeepSwapperMorph
+from facefusion.program_helper import find_argument_group
+from facefusion.thread_helper import thread_semaphore
+from facefusion.types import ApplyStateItem, Args, DownloadScope, Face, InferencePool, Mask, ModelOptions, ModelSet, ProcessMode, QueuePayload, UpdateProgress, VisionFrame
+from facefusion.vision import conditional_match_frame_color, read_image, read_static_image, write_image
+
+
+@lru_cache(maxsize = None)
+def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
+	model_config = []
+
+	if download_scope == 'full':
+		model_config.extend(
+		[
+			('druuzil', 'adam_levine_320'),
+			('druuzil', 'adrianne_palicki_384'),
+			('druuzil', 'agnetha_falskog_224'),
+			('druuzil', 'alan_ritchson_320'),
+			('druuzil', 'alicia_vikander_320'),
+			('druuzil', 'amber_midthunder_320'),
+			('druuzil', 'andras_arato_384'),
+			('druuzil', 'andrew_tate_320'),
+			('druuzil', 'angelina_jolie_384'),
+			('druuzil', 'anne_hathaway_320'),
+			('druuzil', 'anya_chalotra_320'),
+			('druuzil', 'arnold_schwarzenegger_320'),
+			('druuzil', 'benjamin_affleck_320'),
+			('druuzil', 'benjamin_stiller_384'),
+			('druuzil', 'bradley_pitt_224'),
+			('druuzil', 'brie_larson_384'),
+			('druuzil', 'bruce_campbell_384'),
+			('druuzil', 'bryan_cranston_320'),
+			('druuzil', 'catherine_blanchett_352'),
+			('druuzil', 'christian_bale_320'),
+			('druuzil', 'christopher_hemsworth_320'),
+			('druuzil', 'christoph_waltz_384'),
+			('druuzil', 'cillian_murphy_320'),
+			('druuzil', 'cobie_smulders_256'),
+			('druuzil', 'dwayne_johnson_384'),
+			('druuzil', 'edward_norton_320'),
+			('druuzil', 'elisabeth_shue_320'),
+			('druuzil', 'elizabeth_olsen_384'),
+			('druuzil', 'elon_musk_320'),
+			('druuzil', 'emily_blunt_320'),
+			('druuzil', 'emma_stone_384'),
+			('druuzil', 'emma_watson_320'),
+			('druuzil', 'erin_moriarty_384'),
+			('druuzil', 'eva_green_320'),
+			('druuzil', 'ewan_mcgregor_320'),
+			('druuzil', 'florence_pugh_320'),
+			('druuzil', 'freya_allan_320'),
+			('druuzil', 'gary_cole_224'),
+			('druuzil', 'gigi_hadid_224'),
+			('druuzil', 'harrison_ford_384'),
+			('druuzil', 'hayden_christensen_320'),
+			('druuzil', 'heath_ledger_320'),
+			('druuzil', 'henry_cavill_448'),
+			('druuzil', 'hugh_jackman_384'),
+			('druuzil', 'idris_elba_320'),
+			('druuzil', 'jack_nicholson_320'),
+			('druuzil', 'james_carrey_384'),
+			('druuzil', 'james_mcavoy_320'),
+			('druuzil', 'james_varney_320'),
+			('druuzil', 'jason_momoa_320'),
+			('druuzil', 'jason_statham_320'),
+			('druuzil', 'jennifer_connelly_384'),
+			('druuzil', 'jimmy_donaldson_320'),
+			('druuzil', 'jordan_peterson_384'),
+			('druuzil', 'karl_urban_224'),
+			('druuzil', 'kate_beckinsale_384'),
+			('druuzil', 'laurence_fishburne_384'),
+			('druuzil', 'lili_reinhart_320'),
+			('druuzil', 'luke_evans_384'),
+			('druuzil', 'mads_mikkelsen_384'),
+			('druuzil', 'mary_winstead_320'),
+			('druuzil', 'margaret_qualley_384'),
+			('druuzil', 'melina_juergens_320'),
+			('druuzil', 'michael_fassbender_320'),
+			('druuzil', 'michael_fox_320'),
+			('druuzil', 'millie_bobby_brown_320'),
+			('druuzil', 'morgan_freeman_320'),
+			('druuzil', 'patrick_stewart_224'),
+			('druuzil', 'rachel_weisz_384'),
+			('druuzil', 'rebecca_ferguson_320'),
+			('druuzil', 'scarlett_johansson_320'),
+			('druuzil', 'shannen_doherty_384'),
+			('druuzil', 'seth_macfarlane_384'),
+			('druuzil', 'thomas_cruise_320'),
+			('druuzil', 'thomas_hanks_384'),
+			('druuzil', 'william_murray_384'),
+			('druuzil', 'zoe_saldana_384'),
+			('edel', 'emma_roberts_224'),
+			('edel', 'ivanka_trump_224'),
+			('edel', 'lize_dzjabrailova_224'),
+			('edel', 'sidney_sweeney_224'),
+			('edel', 'winona_ryder_224')
+		])
+	if download_scope in [ 'lite', 'full' ]:
+		model_config.extend(
+		[
+			('iperov', 'alexandra_daddario_224'),
+			('iperov', 'alexei_navalny_224'),
+			('iperov', 'amber_heard_224'),
+			('iperov', 'dilraba_dilmurat_224'),
+			('iperov', 'elon_musk_224'),
+			('iperov', 'emilia_clarke_224'),
+			('iperov', 'emma_watson_224'),
+			('iperov', 'erin_moriarty_224'),
+			('iperov', 'jackie_chan_224'),
+			('iperov', 'james_carrey_224'),
+			('iperov', 'jason_statham_320'),
+			('iperov', 'keanu_reeves_320'),
+			('iperov', 'margot_robbie_224'),
+			('iperov', 'natalie_dormer_224'),
+			('iperov', 'nicolas_coppola_224'),
+			('iperov', 'robert_downey_224'),
+			('iperov', 'rowan_atkinson_224'),
+			('iperov', 'ryan_reynolds_224'),
+			('iperov', 'scarlett_johansson_224'),
+			('iperov', 'sylvester_stallone_224'),
+			('iperov', 'thomas_cruise_224'),
+			('iperov', 'thomas_holland_224'),
+			('iperov', 'vin_diesel_224'),
+			('iperov', 'vladimir_putin_224')
+		])
+	if download_scope == 'full':
+		model_config.extend(
+		[
+			('jen', 'angelica_trae_288'),
+			('jen', 'ella_freya_224'),
+			('jen', 'emma_myers_320'),
+			('jen', 'evie_pickerill_224'),
+			('jen', 'kang_hyewon_320'),
+			('jen', 'maddie_mead_224'),
+			('jen', 'nicole_turnbull_288'),
+			('mats', 'alica_schmidt_320'),
+			('mats', 'ashley_alexiss_224'),
+			('mats', 'billie_eilish_224'),
+			('mats', 'brie_larson_224'),
+			('mats', 'cara_delevingne_224'),
+			('mats', 'carolin_kebekus_224'),
+			('mats', 'chelsea_clinton_224'),
+			('mats', 'claire_boucher_224'),
+			('mats', 'corinna_kopf_224'),
+			('mats', 'florence_pugh_224'),
+			('mats', 'hillary_clinton_224'),
+			('mats', 'jenna_fischer_224'),
+			('mats', 'kim_jisoo_320'),
+			('mats', 'mica_suarez_320'),
+			('mats', 'shailene_woodley_224'),
+			('mats', 'shraddha_kapoor_320'),
+			('mats', 'yu_jimin_352'),
+			('rumateus', 'alison_brie_224'),
+			('rumateus', 'amber_heard_224'),
+			('rumateus', 'angelina_jolie_224'),
+			('rumateus', 'aubrey_plaza_224'),
+			('rumateus', 'bridget_regan_224'),
+			('rumateus', 'cobie_smulders_224'),
+			('rumateus', 'deborah_woll_224'),
+			('rumateus', 'dua_lipa_224'),
+			('rumateus', 'emma_stone_224'),
+			('rumateus', 'hailee_steinfeld_224'),
+			('rumateus', 'hilary_duff_224'),
+			('rumateus', 'jessica_alba_224'),
+			('rumateus', 'jessica_biel_224'),
+			('rumateus', 'john_cena_224'),
+			('rumateus', 'kim_kardashian_224'),
+			('rumateus', 'kristen_bell_224'),
+			('rumateus', 'lucy_liu_224'),
+			('rumateus', 'margot_robbie_224'),
+			('rumateus', 'megan_fox_224'),
+			('rumateus', 'meghan_markle_224'),
+			('rumateus', 'millie_bobby_brown_224'),
+			('rumateus', 'natalie_portman_224'),
+			('rumateus', 'nicki_minaj_224'),
+			('rumateus', 'olivia_wilde_224'),
+			('rumateus', 'shay_mitchell_224'),
+			('rumateus', 'sophie_turner_224'),
+			('rumateus', 'taylor_swift_224')
+		])
+	model_set : ModelSet = {}
+
+	for model_scope, model_name in model_config:
+		model_id = '/'.join([ model_scope, model_name ])
+
+		model_set[model_id] =\
+		{
+			'hashes':
+			{
+				'deep_swapper':
+				{
+					'url': resolve_download_url_by_provider('huggingface', 'deepfacelive-models-' + model_scope, model_name + '.hash'),
+					'path': resolve_relative_path('../.assets/models/' + model_scope + '/' + model_name + '.hash')
+				}
+			},
+			'sources':
+			{
+				'deep_swapper':
+				{
+					'url': resolve_download_url_by_provider('huggingface', 'deepfacelive-models-' + model_scope, model_name + '.dfm'),
+					'path': resolve_relative_path('../.assets/models/' + model_scope + '/' + model_name + '.dfm')
+				}
+			},
+			'template': 'dfl_whole_face'
+		}
+
+	custom_model_file_paths = resolve_file_paths(resolve_relative_path('../.assets/models/custom'))
+
+	if custom_model_file_paths:
+
+		for model_file_path in custom_model_file_paths:
+			model_id = '/'.join([ 'custom', get_file_name(model_file_path) ])
+
+			model_set[model_id] =\
+			{
+				'sources':
+				{
+					'deep_swapper':
+					{
+						'path': resolve_relative_path(model_file_path)
+					}
+				},
+				'template': 'dfl_whole_face'
+			}
+
+	return model_set
+
+
+def get_inference_pool() -> InferencePool:
+	model_names = [ state_manager.get_item('deep_swapper_model') ]
+	model_source_set = get_model_options().get('sources')
+
+	return inference_manager.get_inference_pool(__name__, model_names, model_source_set)
+
+
+def clear_inference_pool() -> None:
+	model_names = [ state_manager.get_item('deep_swapper_model') ]
+	inference_manager.clear_inference_pool(__name__, model_names)
+
+
+def get_model_options() -> ModelOptions:
+	model_name = state_manager.get_item('deep_swapper_model')
+	return create_static_model_set('full').get(model_name)
+
+
+def get_model_size() -> Size:
+	deep_swapper = get_inference_pool().get('deep_swapper')
+
+	for deep_swapper_input in deep_swapper.get_inputs():
+		if deep_swapper_input.name == 'in_face:0':
+			return deep_swapper_input.shape[1:3]
+
+	return 0, 0
+
+
+def register_args(program : ArgumentParser) -> None:
+	group_processors = find_argument_group(program, 'processors')
+	if group_processors:
+		group_processors.add_argument('--deep-swapper-model', help = wording.get('help.deep_swapper_model'), default = config.get_str_value('processors', 'deep_swapper_model', 'iperov/elon_musk_224'), choices = processors_choices.deep_swapper_models)
+		group_processors.add_argument('--deep-swapper-morph', help = wording.get('help.deep_swapper_morph'), type = int, default = config.get_int_value('processors', 'deep_swapper_morph', '100'), choices = processors_choices.deep_swapper_morph_range, metavar = create_int_metavar(processors_choices.deep_swapper_morph_range))
+		facefusion.jobs.job_store.register_step_keys([ 'deep_swapper_model', 'deep_swapper_morph' ])
+
+
+def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
+	apply_state_item('deep_swapper_model', args.get('deep_swapper_model'))
+	apply_state_item('deep_swapper_morph', args.get('deep_swapper_morph'))
+
+
+def pre_check() -> bool:
+	model_hash_set = get_model_options().get('hashes')
+	model_source_set = get_model_options().get('sources')
+
+	if model_hash_set and model_source_set:
+		return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)
+	return True
+
+
+def pre_process(mode : ProcessMode) -> bool:
+	if mode in [ 'output', 'preview' ] and not is_image(state_manager.get_item('target_path')) and not is_video(state_manager.get_item('target_path')):
+		logger.error(wording.get('choose_image_or_video_target') + wording.get('exclamation_mark'), __name__)
+		return False
+	if mode == 'output' and not in_directory(state_manager.get_item('output_path')):
+		logger.error(wording.get('specify_image_or_video_output') + wording.get('exclamation_mark'), __name__)
+		return False
+	if mode == 'output' and not same_file_extension(state_manager.get_item('target_path'), state_manager.get_item('output_path')):
+		logger.error(wording.get('match_target_and_output_extension') + wording.get('exclamation_mark'), __name__)
+		return False
+	return True
+
+
+def post_process() -> None:
+	read_static_image.cache_clear()
+	video_manager.clear_video_pool()
+	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
+		clear_inference_pool()
+	if state_manager.get_item('video_memory_strategy') == 'strict':
+		content_analyser.clear_inference_pool()
+		face_classifier.clear_inference_pool()
+		face_detector.clear_inference_pool()
+		face_landmarker.clear_inference_pool()
+		face_masker.clear_inference_pool()
+		face_recognizer.clear_inference_pool()
+
+
+def swap_face(target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
+	model_template = get_model_options().get('template')
+	model_size = get_model_size()
+	crop_vision_frame, affine_matrix = warp_face_by_face_landmark_5(temp_vision_frame, target_face.landmark_set.get('5/68'), model_template, model_size)
+	crop_vision_frame_raw = crop_vision_frame.copy()
+	box_mask = create_box_mask(crop_vision_frame, state_manager.get_item('face_mask_blur'), state_manager.get_item('face_mask_padding'))
+	crop_masks =\
+	[
+		box_mask
+	]
+
+	if 'occlusion' in state_manager.get_item('face_mask_types'):
+		occlusion_mask = create_occlusion_mask(crop_vision_frame)
+		crop_masks.append(occlusion_mask)
+
+	crop_vision_frame = prepare_crop_frame(crop_vision_frame)
+	deep_swapper_morph = numpy.array([ numpy.interp(state_manager.get_item('deep_swapper_morph'), [ 0, 100 ], [ 0, 1 ]) ]).astype(numpy.float32)
+	crop_vision_frame, crop_source_mask, crop_target_mask = forward(crop_vision_frame, deep_swapper_morph)
+	crop_vision_frame = normalize_crop_frame(crop_vision_frame)
+	crop_vision_frame = conditional_match_frame_color(crop_vision_frame_raw, crop_vision_frame)
+	crop_masks.append(prepare_crop_mask(crop_source_mask, crop_target_mask))
+
+	if 'area' in state_manager.get_item('face_mask_types'):
+		face_landmark_68 = cv2.transform(target_face.landmark_set.get('68').reshape(1, -1, 2), affine_matrix).reshape(-1, 2)
+		area_mask = create_area_mask(crop_vision_frame, face_landmark_68, state_manager.get_item('face_mask_areas'))
+		crop_masks.append(area_mask)
+
+	if 'region' in state_manager.get_item('face_mask_types'):
+		region_mask = create_region_mask(crop_vision_frame, state_manager.get_item('face_mask_regions'))
+		crop_masks.append(region_mask)
+
+	crop_mask = numpy.minimum.reduce(crop_masks).clip(0, 1)
+	paste_vision_frame = paste_back(temp_vision_frame, crop_vision_frame, crop_mask, affine_matrix)
+	return paste_vision_frame
+
+
+def forward(crop_vision_frame : VisionFrame, deep_swapper_morph : DeepSwapperMorph) -> Tuple[VisionFrame, Mask, Mask]:
+	deep_swapper = get_inference_pool().get('deep_swapper')
+	deep_swapper_inputs = {}
+
+	for deep_swapper_input in deep_swapper.get_inputs():
+		if deep_swapper_input.name == 'in_face:0':
+			deep_swapper_inputs[deep_swapper_input.name] = crop_vision_frame
+		if deep_swapper_input.name == 'morph_value:0':
+			deep_swapper_inputs[deep_swapper_input.name] = deep_swapper_morph
+
+	with thread_semaphore():
+		crop_target_mask, crop_vision_frame, crop_source_mask = deep_swapper.run(None, deep_swapper_inputs)
+
+	return crop_vision_frame[0], crop_source_mask[0], crop_target_mask[0]
+
+
+def has_morph_input() -> bool:
+	deep_swapper = get_inference_pool().get('deep_swapper')
+
+	for deep_swapper_input in deep_swapper.get_inputs():
+		if deep_swapper_input.name == 'morph_value:0':
+			return True
+
+	return False
+
+
+def prepare_crop_frame(crop_vision_frame : VisionFrame) -> VisionFrame:
+	crop_vision_frame = cv2.addWeighted(crop_vision_frame, 1.75, cv2.GaussianBlur(crop_vision_frame, (0, 0), 2), -0.75, 0)
+	crop_vision_frame = crop_vision_frame / 255.0
+	crop_vision_frame = numpy.expand_dims(crop_vision_frame, axis = 0).astype(numpy.float32)
+	return crop_vision_frame
+
+
+def normalize_crop_frame(crop_vision_frame : VisionFrame) -> VisionFrame:
+	crop_vision_frame = (crop_vision_frame * 255.0).clip(0, 255)
+	crop_vision_frame = crop_vision_frame.astype(numpy.uint8)
+	return crop_vision_frame
+
+
+def prepare_crop_mask(crop_source_mask : Mask, crop_target_mask : Mask) -> Mask:
+	model_size = get_model_size()
+	blur_size = 6.25
+	kernel_size = 3
+	crop_mask = numpy.minimum.reduce([ crop_source_mask, crop_target_mask ])
+	crop_mask = crop_mask.reshape(model_size).clip(0, 1)
+	crop_mask = cv2.erode(crop_mask, cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (kernel_size, kernel_size)), iterations = 2)
+	crop_mask = cv2.GaussianBlur(crop_mask, (0, 0), blur_size)
+	return crop_mask
+
+
+def get_reference_frame(source_face : Face, target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
+	return swap_face(target_face, temp_vision_frame)
+
+
+def process_frame(inputs : DeepSwapperInputs) -> VisionFrame:
+	reference_faces = inputs.get('reference_faces')
+	target_vision_frame = inputs.get('target_vision_frame')
+	many_faces = sort_and_filter_faces(get_many_faces([ target_vision_frame ]))
+
+	if state_manager.get_item('face_selector_mode') == 'many':
+		if many_faces:
+			for target_face in many_faces:
+				target_vision_frame = swap_face(target_face, target_vision_frame)
+	if state_manager.get_item('face_selector_mode') == 'one':
+		target_face = get_one_face(many_faces)
+		if target_face:
+			target_vision_frame = swap_face(target_face, target_vision_frame)
+	if state_manager.get_item('face_selector_mode') == 'reference':
+		similar_faces = find_similar_faces(many_faces, reference_faces, state_manager.get_item('reference_face_distance'))
+		if similar_faces:
+			for similar_face in similar_faces:
+				target_vision_frame = swap_face(similar_face, target_vision_frame)
+	return target_vision_frame
+
+
+def process_frames(source_path : List[str], queue_payloads : List[QueuePayload], update_progress : UpdateProgress) -> None:
+	reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
+
+	for queue_payload in process_manager.manage(queue_payloads):
+		target_vision_path = queue_payload['frame_path']
+		target_vision_frame = read_image(target_vision_path)
+		output_vision_frame = process_frame(
+		{
+			'reference_faces': reference_faces,
+			'target_vision_frame': target_vision_frame
+		})
+		write_image(target_vision_path, output_vision_frame)
+		update_progress(1)
+
+
+def process_image(source_path : str, target_path : str, output_path : str) -> None:
+	reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
+	target_vision_frame = read_static_image(target_path)
+	output_vision_frame = process_frame(
+	{
+		'reference_faces': reference_faces,
+		'target_vision_frame': target_vision_frame
+	})
+	write_image(output_path, output_vision_frame)
+
+
+def process_video(source_paths : List[str], temp_frame_paths : List[str]) -> None:
+	processors.multi_process_frames(None, temp_frame_paths, process_frames)

facefusion/processors/modules/expression_restorer.py

@@ -0,0 +1,298 @@
+from argparse import ArgumentParser
+from functools import lru_cache
+from typing import List, Tuple
+
+import cv2
+import numpy
+
+import facefusion.jobs.job_manager
+import facefusion.jobs.job_store
+import facefusion.processors.core as processors
+from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, inference_manager, logger, process_manager, state_manager, video_manager, wording
+from facefusion.common_helper import create_int_metavar
+from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
+from facefusion.face_analyser import get_many_faces, get_one_face
+from facefusion.face_helper import paste_back, warp_face_by_face_landmark_5
+from facefusion.face_masker import create_box_mask, create_occlusion_mask
+from facefusion.face_selector import find_similar_faces, sort_and_filter_faces
+from facefusion.face_store import get_reference_faces
+from facefusion.filesystem import in_directory, is_image, is_video, resolve_relative_path, same_file_extension
+from facefusion.processors import choices as processors_choices
+from facefusion.processors.live_portrait import create_rotation, limit_expression
+from facefusion.processors.types import ExpressionRestorerInputs, LivePortraitExpression, LivePortraitFeatureVolume, LivePortraitMotionPoints, LivePortraitPitch, LivePortraitRoll, LivePortraitScale, LivePortraitTranslation, LivePortraitYaw
+from facefusion.program_helper import find_argument_group
+from facefusion.thread_helper import conditional_thread_semaphore, thread_semaphore
+from facefusion.types import ApplyStateItem, Args, DownloadScope, Face, InferencePool, ModelOptions, ModelSet, ProcessMode, QueuePayload, UpdateProgress, VisionFrame
+from facefusion.vision import read_image, read_static_image, read_video_frame, write_image
+
+
+@lru_cache(maxsize = None)
+def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
+	return\
+	{
+		'live_portrait':
+		{
+			'hashes':
+			{
+				'feature_extractor':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_feature_extractor.hash'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_feature_extractor.hash')
+				},
+				'motion_extractor':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_motion_extractor.hash'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_motion_extractor.hash')
+				},
+				'generator':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_generator.hash'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_generator.hash')
+				}
+			},
+			'sources':
+			{
+				'feature_extractor':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_feature_extractor.onnx'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_feature_extractor.onnx')
+				},
+				'motion_extractor':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_motion_extractor.onnx'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_motion_extractor.onnx')
+				},
+				'generator':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_generator.onnx'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_generator.onnx')
+				}
+			},
+			'template': 'arcface_128',
+			'size': (512, 512)
+		}
+	}
+
+
+def get_inference_pool() -> InferencePool:
+	model_names = [ state_manager.get_item('expression_restorer_model') ]
+	model_source_set = get_model_options().get('sources')
+
+	return inference_manager.get_inference_pool(__name__, model_names, model_source_set)
+
+
+def clear_inference_pool() -> None:
+	model_names = [ state_manager.get_item('expression_restorer_model') ]
+	inference_manager.clear_inference_pool(__name__, model_names)
+
+
+def get_model_options() -> ModelOptions:
+	model_name = state_manager.get_item('expression_restorer_model')
+	return create_static_model_set('full').get(model_name)
+
+
+def register_args(program : ArgumentParser) -> None:
+	group_processors = find_argument_group(program, 'processors')
+	if group_processors:
+		group_processors.add_argument('--expression-restorer-model', help = wording.get('help.expression_restorer_model'), default = config.get_str_value('processors', 'expression_restorer_model', 'live_portrait'), choices = processors_choices.expression_restorer_models)
+		group_processors.add_argument('--expression-restorer-factor', help = wording.get('help.expression_restorer_factor'), type = int, default = config.get_int_value('processors', 'expression_restorer_factor', '80'), choices = processors_choices.expression_restorer_factor_range, metavar = create_int_metavar(processors_choices.expression_restorer_factor_range))
+		facefusion.jobs.job_store.register_step_keys([ 'expression_restorer_model', 'expression_restorer_factor' ])
+
+
+def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
+	apply_state_item('expression_restorer_model', args.get('expression_restorer_model'))
+	apply_state_item('expression_restorer_factor', args.get('expression_restorer_factor'))
+
+
+def pre_check() -> bool:
+	model_hash_set = get_model_options().get('hashes')
+	model_source_set = get_model_options().get('sources')
+
+	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)
+
+
+def pre_process(mode : ProcessMode) -> bool:
+	if mode == 'stream':
+		logger.error(wording.get('stream_not_supported') + wording.get('exclamation_mark'), __name__)
+		return False
+	if mode in [ 'output', 'preview' ] and not is_image(state_manager.get_item('target_path')) and not is_video(state_manager.get_item('target_path')):
+		logger.error(wording.get('choose_image_or_video_target') + wording.get('exclamation_mark'), __name__)
+		return False
+	if mode == 'output' and not in_directory(state_manager.get_item('output_path')):
+		logger.error(wording.get('specify_image_or_video_output') + wording.get('exclamation_mark'), __name__)
+		return False
+	if mode == 'output' and not same_file_extension(state_manager.get_item('target_path'), state_manager.get_item('output_path')):
+		logger.error(wording.get('match_target_and_output_extension') + wording.get('exclamation_mark'), __name__)
+		return False
+	return True
+
+
+def post_process() -> None:
+	read_static_image.cache_clear()
+	video_manager.clear_video_pool()
+	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
+		clear_inference_pool()
+	if state_manager.get_item('video_memory_strategy') == 'strict':
+		content_analyser.clear_inference_pool()
+		face_classifier.clear_inference_pool()
+		face_detector.clear_inference_pool()
+		face_landmarker.clear_inference_pool()
+		face_masker.clear_inference_pool()
+		face_recognizer.clear_inference_pool()
+
+
+def restore_expression(source_vision_frame : VisionFrame, target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
+	model_template = get_model_options().get('template')
+	model_size = get_model_options().get('size')
+	expression_restorer_factor = float(numpy.interp(float(state_manager.get_item('expression_restorer_factor')), [ 0, 100 ], [ 0, 1.2 ]))
+	source_vision_frame = cv2.resize(source_vision_frame, temp_vision_frame.shape[:2][::-1])
+	source_crop_vision_frame, _ = warp_face_by_face_landmark_5(source_vision_frame, target_face.landmark_set.get('5/68'), model_template, model_size)
+	target_crop_vision_frame, affine_matrix = warp_face_by_face_landmark_5(temp_vision_frame, target_face.landmark_set.get('5/68'), model_template, model_size)
+	box_mask = create_box_mask(target_crop_vision_frame, state_manager.get_item('face_mask_blur'), (0, 0, 0, 0))
+	crop_masks =\
+	[
+		box_mask
+	]
+
+	if 'occlusion' in state_manager.get_item('face_mask_types'):
+		occlusion_mask = create_occlusion_mask(target_crop_vision_frame)
+		crop_masks.append(occlusion_mask)
+
+	source_crop_vision_frame = prepare_crop_frame(source_crop_vision_frame)
+	target_crop_vision_frame = prepare_crop_frame(target_crop_vision_frame)
+	target_crop_vision_frame = apply_restore(source_crop_vision_frame, target_crop_vision_frame, expression_restorer_factor)
+	target_crop_vision_frame = normalize_crop_frame(target_crop_vision_frame)
+	crop_mask = numpy.minimum.reduce(crop_masks).clip(0, 1)
+	temp_vision_frame = paste_back(temp_vision_frame, target_crop_vision_frame, crop_mask, affine_matrix)
+	return temp_vision_frame
+
+
+def apply_restore(source_crop_vision_frame : VisionFrame, target_crop_vision_frame : VisionFrame, expression_restorer_factor : float) -> VisionFrame:
+	feature_volume = forward_extract_feature(target_crop_vision_frame)
+	source_expression = forward_extract_motion(source_crop_vision_frame)[5]
+	pitch, yaw, roll, scale, translation, target_expression, motion_points = forward_extract_motion(target_crop_vision_frame)
+	rotation = create_rotation(pitch, yaw, roll)
+	source_expression[:, [ 0, 4, 5, 8, 9 ]] = target_expression[:, [ 0, 4, 5, 8, 9 ]]
+	source_expression = source_expression * expression_restorer_factor + target_expression * (1 - expression_restorer_factor)
+	source_expression = limit_expression(source_expression)
+	source_motion_points = scale * (motion_points @ rotation.T + source_expression) + translation
+	target_motion_points = scale * (motion_points @ rotation.T + target_expression) + translation
+	crop_vision_frame = forward_generate_frame(feature_volume, source_motion_points, target_motion_points)
+	return crop_vision_frame
+
+
+def forward_extract_feature(crop_vision_frame : VisionFrame) -> LivePortraitFeatureVolume:
+	feature_extractor = get_inference_pool().get('feature_extractor')
+
+	with conditional_thread_semaphore():
+		feature_volume = feature_extractor.run(None,
+		{
+			'input': crop_vision_frame
+		})[0]
+
+	return feature_volume
+
+
+def forward_extract_motion(crop_vision_frame : VisionFrame) -> Tuple[LivePortraitPitch, LivePortraitYaw, LivePortraitRoll, LivePortraitScale, LivePortraitTranslation, LivePortraitExpression, LivePortraitMotionPoints]:
+	motion_extractor = get_inference_pool().get('motion_extractor')
+
+	with conditional_thread_semaphore():
+		pitch, yaw, roll, scale, translation, expression, motion_points = motion_extractor.run(None,
+		{
+			'input': crop_vision_frame
+		})
+
+	return pitch, yaw, roll, scale, translation, expression, motion_points
+
+
+def forward_generate_frame(feature_volume : LivePortraitFeatureVolume, source_motion_points : LivePortraitMotionPoints, target_motion_points : LivePortraitMotionPoints) -> VisionFrame:
+	generator = get_inference_pool().get('generator')
+
+	with thread_semaphore():
+		crop_vision_frame = generator.run(None,
+		{
+			'feature_volume': feature_volume,
+			'source': source_motion_points,
+			'target': target_motion_points
+		})[0][0]
+
+	return crop_vision_frame
+
+
+def prepare_crop_frame(crop_vision_frame : VisionFrame) -> VisionFrame:
+	model_size = get_model_options().get('size')
+	prepare_size = (model_size[0] // 2, model_size[1] // 2)
+	crop_vision_frame = cv2.resize(crop_vision_frame, prepare_size, interpolation = cv2.INTER_AREA)
+	crop_vision_frame = crop_vision_frame[:, :, ::-1] / 255.0
+	crop_vision_frame = numpy.expand_dims(crop_vision_frame.transpose(2, 0, 1), axis = 0).astype(numpy.float32)
+	return crop_vision_frame
+
+
+def normalize_crop_frame(crop_vision_frame : VisionFrame) -> VisionFrame:
+	crop_vision_frame = crop_vision_frame.transpose(1, 2, 0).clip(0, 1)
+	crop_vision_frame = crop_vision_frame * 255.0
+	crop_vision_frame = crop_vision_frame.astype(numpy.uint8)[:, :, ::-1]
+	return crop_vision_frame
+
+
+def get_reference_frame(source_face : Face, target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
+	pass
+
+
+def process_frame(inputs : ExpressionRestorerInputs) -> VisionFrame:
+	reference_faces = inputs.get('reference_faces')
+	source_vision_frame = inputs.get('source_vision_frame')
+	target_vision_frame = inputs.get('target_vision_frame')
+	many_faces = sort_and_filter_faces(get_many_faces([ target_vision_frame ]))
+
+	if state_manager.get_item('face_selector_mode') == 'many':
+		if many_faces:
+			for target_face in many_faces:
+				target_vision_frame = restore_expression(source_vision_frame, target_face, target_vision_frame)
+	if state_manager.get_item('face_selector_mode') == 'one':
+		target_face = get_one_face(many_faces)
+		if target_face:
+			target_vision_frame = restore_expression(source_vision_frame, target_face, target_vision_frame)
+	if state_manager.get_item('face_selector_mode') == 'reference':
+		similar_faces = find_similar_faces(many_faces, reference_faces, state_manager.get_item('reference_face_distance'))
+		if similar_faces:
+			for similar_face in similar_faces:
+				target_vision_frame = restore_expression(source_vision_frame, similar_face, target_vision_frame)
+	return target_vision_frame
+
+
+def process_frames(source_path : List[str], queue_payloads : List[QueuePayload], update_progress : UpdateProgress) -> None:
+	reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
+
+	for queue_payload in process_manager.manage(queue_payloads):
+		frame_number = queue_payload.get('frame_number')
+		if state_manager.get_item('trim_frame_start'):
+			frame_number += state_manager.get_item('trim_frame_start')
+		source_vision_frame = read_video_frame(state_manager.get_item('target_path'), frame_number)
+		target_vision_path = queue_payload.get('frame_path')
+		target_vision_frame = read_image(target_vision_path)
+		output_vision_frame = process_frame(
+		{
+			'reference_faces': reference_faces,
+			'source_vision_frame': source_vision_frame,
+			'target_vision_frame': target_vision_frame
+		})
+		write_image(target_vision_path, output_vision_frame)
+		update_progress(1)
+
+
+def process_image(source_path : str, target_path : str, output_path : str) -> None:
+	reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
+	source_vision_frame = read_static_image(state_manager.get_item('target_path'))
+	target_vision_frame = read_static_image(target_path)
+	output_vision_frame = process_frame(
+	{
+		'reference_faces': reference_faces,
+		'source_vision_frame': source_vision_frame,
+		'target_vision_frame': target_vision_frame
+	})
+	write_image(output_path, output_vision_frame)
+
+
+def process_video(source_paths : List[str], temp_frame_paths : List[str]) -> None:
+	processors.multi_process_frames(None, temp_frame_paths, process_frames)

facefusion/processors/modules/face_debugger.py

@@ -0,0 +1,228 @@
+from argparse import ArgumentParser
+from typing import List
+
+import cv2
+import numpy
+
+import facefusion.jobs.job_manager
+import facefusion.jobs.job_store
+import facefusion.processors.core as processors
+from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, logger, process_manager, state_manager, video_manager, wording
+from facefusion.face_analyser import get_many_faces, get_one_face
+from facefusion.face_helper import warp_face_by_face_landmark_5
+from facefusion.face_masker import create_area_mask, create_box_mask, create_occlusion_mask, create_region_mask
+from facefusion.face_selector import find_similar_faces, sort_and_filter_faces
+from facefusion.face_store import get_reference_faces
+from facefusion.filesystem import in_directory, same_file_extension
+from facefusion.processors import choices as processors_choices
+from facefusion.processors.types import FaceDebuggerInputs
+from facefusion.program_helper import find_argument_group
+from facefusion.types import ApplyStateItem, Args, Face, InferencePool, ProcessMode, QueuePayload, UpdateProgress, VisionFrame
+from facefusion.vision import read_image, read_static_image, write_image
+
+
+def get_inference_pool() -> InferencePool:
+	pass
+
+
+def clear_inference_pool() -> None:
+	pass
+
+
+def register_args(program : ArgumentParser) -> None:
+	group_processors = find_argument_group(program, 'processors')
+	if group_processors:
+		group_processors.add_argument('--face-debugger-items', help = wording.get('help.face_debugger_items').format(choices = ', '.join(processors_choices.face_debugger_items)), default = config.get_str_list('processors', 'face_debugger_items', 'face-landmark-5/68 face-mask'), choices = processors_choices.face_debugger_items, nargs = '+', metavar = 'FACE_DEBUGGER_ITEMS')
+		facefusion.jobs.job_store.register_step_keys([ 'face_debugger_items' ])
+
+
+def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
+	apply_state_item('face_debugger_items', args.get('face_debugger_items'))
+
+
+def pre_check() -> bool:
+	return True
+
+
+def pre_process(mode : ProcessMode) -> bool:
+	if mode == 'output' and not in_directory(state_manager.get_item('output_path')):
+		logger.error(wording.get('specify_image_or_video_output') + wording.get('exclamation_mark'), __name__)
+		return False
+	if mode == 'output' and not same_file_extension(state_manager.get_item('target_path'), state_manager.get_item('output_path')):
+		logger.error(wording.get('match_target_and_output_extension') + wording.get('exclamation_mark'), __name__)
+		return False
+	return True
+
+
+def post_process() -> None:
+	read_static_image.cache_clear()
+	video_manager.clear_video_pool()
+	if state_manager.get_item('video_memory_strategy') == 'strict':
+		content_analyser.clear_inference_pool()
+		face_classifier.clear_inference_pool()
+		face_detector.clear_inference_pool()
+		face_landmarker.clear_inference_pool()
+		face_masker.clear_inference_pool()
+		face_recognizer.clear_inference_pool()
+
+
+def debug_face(target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
+	primary_color = (0, 0, 255)
+	primary_light_color = (100, 100, 255)
+	secondary_color = (0, 255, 0)
+	tertiary_color = (255, 255, 0)
+	bounding_box = target_face.bounding_box.astype(numpy.int32)
+	temp_vision_frame = temp_vision_frame.copy()
+	has_face_landmark_5_fallback = numpy.array_equal(target_face.landmark_set.get('5'), target_face.landmark_set.get('5/68'))
+	has_face_landmark_68_fallback = numpy.array_equal(target_face.landmark_set.get('68'), target_face.landmark_set.get('68/5'))
+	face_debugger_items = state_manager.get_item('face_debugger_items')
+
+	if 'bounding-box' in face_debugger_items:
+		x1, y1, x2, y2 = bounding_box
+		cv2.rectangle(temp_vision_frame, (x1, y1), (x2, y2), primary_color, 2)
+
+		if target_face.angle == 0:
+			cv2.line(temp_vision_frame, (x1, y1), (x2, y1), primary_light_color, 3)
+		if target_face.angle == 180:
+			cv2.line(temp_vision_frame, (x1, y2), (x2, y2), primary_light_color, 3)
+		if target_face.angle == 90:
+			cv2.line(temp_vision_frame, (x2, y1), (x2, y2), primary_light_color, 3)
+		if target_face.angle == 270:
+			cv2.line(temp_vision_frame, (x1, y1), (x1, y2), primary_light_color, 3)
+
+	if 'face-mask' in face_debugger_items:
+		crop_vision_frame, affine_matrix = warp_face_by_face_landmark_5(temp_vision_frame, target_face.landmark_set.get('5/68'), 'arcface_128', (512, 512))
+		inverse_matrix = cv2.invertAffineTransform(affine_matrix)
+		temp_size = temp_vision_frame.shape[:2][::-1]
+		crop_masks = []
+
+		if 'box' in state_manager.get_item('face_mask_types'):
+			box_mask = create_box_mask(crop_vision_frame, 0, state_manager.get_item('face_mask_padding'))
+			crop_masks.append(box_mask)
+
+		if 'occlusion' in state_manager.get_item('face_mask_types'):
+			occlusion_mask = create_occlusion_mask(crop_vision_frame)
+			crop_masks.append(occlusion_mask)
+
+		if 'area' in state_manager.get_item('face_mask_types'):
+			face_landmark_68 = cv2.transform(target_face.landmark_set.get('68').reshape(1, -1, 2), affine_matrix).reshape(-1, 2)
+			area_mask = create_area_mask(crop_vision_frame, face_landmark_68, state_manager.get_item('face_mask_areas'))
+			crop_masks.append(area_mask)
+
+		if 'region' in state_manager.get_item('face_mask_types'):
+			region_mask = create_region_mask(crop_vision_frame, state_manager.get_item('face_mask_regions'))
+			crop_masks.append(region_mask)
+
+		crop_mask = numpy.minimum.reduce(crop_masks).clip(0, 1)
+		crop_mask = (crop_mask * 255).astype(numpy.uint8)
+		inverse_vision_frame = cv2.warpAffine(crop_mask, inverse_matrix, temp_size)
+		inverse_vision_frame = cv2.threshold(inverse_vision_frame, 100, 255, cv2.THRESH_BINARY)[1]
+		inverse_vision_frame[inverse_vision_frame > 0] = 255 #type:ignore[operator]
+		inverse_contours = cv2.findContours(inverse_vision_frame, cv2.RETR_LIST, cv2.CHAIN_APPROX_NONE)[0]
+		cv2.drawContours(temp_vision_frame, inverse_contours, -1, tertiary_color if has_face_landmark_5_fallback else secondary_color, 2)
+
+	if 'face-landmark-5' in face_debugger_items and numpy.any(target_face.landmark_set.get('5')):
+		face_landmark_5 = target_face.landmark_set.get('5').astype(numpy.int32)
+		for index in range(face_landmark_5.shape[0]):
+			cv2.circle(temp_vision_frame, (face_landmark_5[index][0], face_landmark_5[index][1]), 3, primary_color, -1)
+
+	if 'face-landmark-5/68' in face_debugger_items and numpy.any(target_face.landmark_set.get('5/68')):
+		face_landmark_5_68 = target_face.landmark_set.get('5/68').astype(numpy.int32)
+		for index in range(face_landmark_5_68.shape[0]):
+			cv2.circle(temp_vision_frame, (face_landmark_5_68[index][0], face_landmark_5_68[index][1]), 3, tertiary_color if has_face_landmark_5_fallback else secondary_color, -1)
+
+	if 'face-landmark-68' in face_debugger_items and numpy.any(target_face.landmark_set.get('68')):
+		face_landmark_68 = target_face.landmark_set.get('68').astype(numpy.int32)
+		for index in range(face_landmark_68.shape[0]):
+			cv2.circle(temp_vision_frame, (face_landmark_68[index][0], face_landmark_68[index][1]), 3, tertiary_color if has_face_landmark_68_fallback else secondary_color, -1)
+
+	if 'face-landmark-68/5' in face_debugger_items and numpy.any(target_face.landmark_set.get('68')):
+		face_landmark_68 = target_face.landmark_set.get('68/5').astype(numpy.int32)
+		for index in range(face_landmark_68.shape[0]):
+			cv2.circle(temp_vision_frame, (face_landmark_68[index][0], face_landmark_68[index][1]), 3, tertiary_color, -1)
+
+	if bounding_box[3] - bounding_box[1] > 50 and bounding_box[2] - bounding_box[0] > 50:
+		top = bounding_box[1]
+		left = bounding_box[0] - 20
+
+		if 'face-detector-score' in face_debugger_items:
+			face_score_text = str(round(target_face.score_set.get('detector'), 2))
+			top = top + 20
+			cv2.putText(temp_vision_frame, face_score_text, (left, top), cv2.FONT_HERSHEY_SIMPLEX, 0.5, primary_color, 2)
+
+		if 'face-landmarker-score' in face_debugger_items:
+			face_score_text = str(round(target_face.score_set.get('landmarker'), 2))
+			top = top + 20
+			cv2.putText(temp_vision_frame, face_score_text, (left, top), cv2.FONT_HERSHEY_SIMPLEX, 0.5, tertiary_color if has_face_landmark_5_fallback else secondary_color, 2)
+
+		if 'age' in face_debugger_items:
+			face_age_text = str(target_face.age.start) + '-' + str(target_face.age.stop)
+			top = top + 20
+			cv2.putText(temp_vision_frame, face_age_text, (left, top), cv2.FONT_HERSHEY_SIMPLEX, 0.5, primary_color, 2)
+
+		if 'gender' in face_debugger_items:
+			face_gender_text = target_face.gender
+			top = top + 20
+			cv2.putText(temp_vision_frame, face_gender_text, (left, top), cv2.FONT_HERSHEY_SIMPLEX, 0.5, primary_color, 2)
+
+		if 'race' in face_debugger_items:
+			face_race_text = target_face.race
+			top = top + 20
+			cv2.putText(temp_vision_frame, face_race_text, (left, top), cv2.FONT_HERSHEY_SIMPLEX, 0.5, primary_color, 2)
+
+	return temp_vision_frame
+
+
+def get_reference_frame(source_face : Face, target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
+	pass
+
+
+def process_frame(inputs : FaceDebuggerInputs) -> VisionFrame:
+	reference_faces = inputs.get('reference_faces')
+	target_vision_frame = inputs.get('target_vision_frame')
+	many_faces = sort_and_filter_faces(get_many_faces([ target_vision_frame ]))
+
+	if state_manager.get_item('face_selector_mode') == 'many':
+		if many_faces:
+			for target_face in many_faces:
+				target_vision_frame = debug_face(target_face, target_vision_frame)
+	if state_manager.get_item('face_selector_mode') == 'one':
+		target_face = get_one_face(many_faces)
+		if target_face:
+			target_vision_frame = debug_face(target_face, target_vision_frame)
+	if state_manager.get_item('face_selector_mode') == 'reference':
+		similar_faces = find_similar_faces(many_faces, reference_faces, state_manager.get_item('reference_face_distance'))
+		if similar_faces:
+			for similar_face in similar_faces:
+				target_vision_frame = debug_face(similar_face, target_vision_frame)
+	return target_vision_frame
+
+
+def process_frames(source_paths : List[str], queue_payloads : List[QueuePayload], update_progress : UpdateProgress) -> None:
+	reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
+
+	for queue_payload in process_manager.manage(queue_payloads):
+		target_vision_path = queue_payload['frame_path']
+		target_vision_frame = read_image(target_vision_path)
+		output_vision_frame = process_frame(
+		{
+			'reference_faces': reference_faces,
+			'target_vision_frame': target_vision_frame
+		})
+		write_image(target_vision_path, output_vision_frame)
+		update_progress(1)
+
+
+def process_image(source_paths : List[str], target_path : str, output_path : str) -> None:
+	reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
+	target_vision_frame = read_static_image(target_path)
+	output_vision_frame = process_frame(
+	{
+		'reference_faces': reference_faces,
+		'target_vision_frame': target_vision_frame
+	})
+	write_image(output_path, output_vision_frame)
+
+
+def process_video(source_paths : List[str], temp_frame_paths : List[str]) -> None:
+	processors.multi_process_frames(source_paths, temp_frame_paths, process_frames)

facefusion/processors/modules/face_editor.py

@@ -0,0 +1,533 @@
+from argparse import ArgumentParser
+from functools import lru_cache
+from typing import List, Tuple
+
+import cv2
+import numpy
+
+import facefusion.jobs.job_manager
+import facefusion.jobs.job_store
+import facefusion.processors.core as processors
+from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, inference_manager, logger, process_manager, state_manager, video_manager, wording
+from facefusion.common_helper import create_float_metavar
+from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
+from facefusion.face_analyser import get_many_faces, get_one_face
+from facefusion.face_helper import paste_back, scale_face_landmark_5, warp_face_by_face_landmark_5
+from facefusion.face_masker import create_box_mask
+from facefusion.face_selector import find_similar_faces, sort_and_filter_faces
+from facefusion.face_store import get_reference_faces
+from facefusion.filesystem import in_directory, is_image, is_video, resolve_relative_path, same_file_extension
+from facefusion.processors import choices as processors_choices
+from facefusion.processors.live_portrait import create_rotation, limit_euler_angles, limit_expression
+from facefusion.processors.types import FaceEditorInputs, LivePortraitExpression, LivePortraitFeatureVolume, LivePortraitMotionPoints, LivePortraitPitch, LivePortraitRoll, LivePortraitRotation, LivePortraitScale, LivePortraitTranslation, LivePortraitYaw
+from facefusion.program_helper import find_argument_group
+from facefusion.thread_helper import conditional_thread_semaphore, thread_semaphore
+from facefusion.types import ApplyStateItem, Args, DownloadScope, Face, FaceLandmark68, InferencePool, ModelOptions, ModelSet, ProcessMode, QueuePayload, UpdateProgress, VisionFrame
+from facefusion.vision import read_image, read_static_image, write_image
+
+
+@lru_cache(maxsize = None)
+def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
+	return\
+	{
+		'live_portrait':
+		{
+			'hashes':
+			{
+				'feature_extractor':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_feature_extractor.hash'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_feature_extractor.hash')
+				},
+				'motion_extractor':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_motion_extractor.hash'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_motion_extractor.hash')
+				},
+				'eye_retargeter':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_eye_retargeter.hash'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_eye_retargeter.hash')
+				},
+				'lip_retargeter':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_lip_retargeter.hash'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_lip_retargeter.hash')
+				},
+				'stitcher':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_stitcher.hash'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_stitcher.hash')
+				},
+				'generator':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_generator.hash'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_generator.hash')
+				}
+			},
+			'sources':
+			{
+				'feature_extractor':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_feature_extractor.onnx'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_feature_extractor.onnx')
+				},
+				'motion_extractor':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_motion_extractor.onnx'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_motion_extractor.onnx')
+				},
+				'eye_retargeter':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_eye_retargeter.onnx'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_eye_retargeter.onnx')
+				},
+				'lip_retargeter':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_lip_retargeter.onnx'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_lip_retargeter.onnx')
+				},
+				'stitcher':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_stitcher.onnx'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_stitcher.onnx')
+				},
+				'generator':
+				{
+					'url': resolve_download_url('models-3.0.0', 'live_portrait_generator.onnx'),
+					'path': resolve_relative_path('../.assets/models/live_portrait_generator.onnx')
+				}
+			},
+			'template': 'ffhq_512',
+			'size': (512, 512)
+		}
+	}
+
+
+def get_inference_pool() -> InferencePool:
+	model_names = [ state_manager.get_item('face_editor_model') ]
+	model_source_set = get_model_options().get('sources')
+
+	return inference_manager.get_inference_pool(__name__, model_names, model_source_set)
+
+
+def clear_inference_pool() -> None:
+	model_names = [ state_manager.get_item('face_editor_model') ]
+	inference_manager.clear_inference_pool(__name__, model_names)
+
+
+def get_model_options() -> ModelOptions:
+	model_name = state_manager.get_item('face_editor_model')
+	return create_static_model_set('full').get(model_name)
+
+
+def register_args(program : ArgumentParser) -> None:
+	group_processors = find_argument_group(program, 'processors')
+	if group_processors:
+		group_processors.add_argument('--face-editor-model', help = wording.get('help.face_editor_model'), default = config.get_str_value('processors', 'face_editor_model', 'live_portrait'), choices = processors_choices.face_editor_models)
+		group_processors.add_argument('--face-editor-eyebrow-direction', help = wording.get('help.face_editor_eyebrow_direction'), type = float, default = config.get_float_value('processors', 'face_editor_eyebrow_direction', '0'), choices = processors_choices.face_editor_eyebrow_direction_range, metavar = create_float_metavar(processors_choices.face_editor_eyebrow_direction_range))
+		group_processors.add_argument('--face-editor-eye-gaze-horizontal', help = wording.get('help.face_editor_eye_gaze_horizontal'), type = float, default = config.get_float_value('processors', 'face_editor_eye_gaze_horizontal', '0'), choices = processors_choices.face_editor_eye_gaze_horizontal_range, metavar = create_float_metavar(processors_choices.face_editor_eye_gaze_horizontal_range))
+		group_processors.add_argument('--face-editor-eye-gaze-vertical', help = wording.get('help.face_editor_eye_gaze_vertical'), type = float, default = config.get_float_value('processors', 'face_editor_eye_gaze_vertical', '0'), choices = processors_choices.face_editor_eye_gaze_vertical_range, metavar = create_float_metavar(processors_choices.face_editor_eye_gaze_vertical_range))
+		group_processors.add_argument('--face-editor-eye-open-ratio', help = wording.get('help.face_editor_eye_open_ratio'), type = float, default = config.get_float_value('processors', 'face_editor_eye_open_ratio', '0'), choices = processors_choices.face_editor_eye_open_ratio_range, metavar = create_float_metavar(processors_choices.face_editor_eye_open_ratio_range))
+		group_processors.add_argument('--face-editor-lip-open-ratio', help = wording.get('help.face_editor_lip_open_ratio'), type = float, default = config.get_float_value('processors', 'face_editor_lip_open_ratio', '0'), choices = processors_choices.face_editor_lip_open_ratio_range, metavar = create_float_metavar(processors_choices.face_editor_lip_open_ratio_range))
+		group_processors.add_argument('--face-editor-mouth-grim', help = wording.get('help.face_editor_mouth_grim'), type = float, default = config.get_float_value('processors', 'face_editor_mouth_grim', '0'), choices = processors_choices.face_editor_mouth_grim_range, metavar = create_float_metavar(processors_choices.face_editor_mouth_grim_range))
+		group_processors.add_argument('--face-editor-mouth-pout', help = wording.get('help.face_editor_mouth_pout'), type = float, default = config.get_float_value('processors', 'face_editor_mouth_pout', '0'), choices = processors_choices.face_editor_mouth_pout_range, metavar = create_float_metavar(processors_choices.face_editor_mouth_pout_range))
+		group_processors.add_argument('--face-editor-mouth-purse', help = wording.get('help.face_editor_mouth_purse'), type = float, default = config.get_float_value('processors', 'face_editor_mouth_purse', '0'), choices = processors_choices.face_editor_mouth_purse_range, metavar = create_float_metavar(processors_choices.face_editor_mouth_purse_range))
+		group_processors.add_argument('--face-editor-mouth-smile', help = wording.get('help.face_editor_mouth_smile'), type = float, default = config.get_float_value('processors', 'face_editor_mouth_smile', '0'), choices = processors_choices.face_editor_mouth_smile_range, metavar = create_float_metavar(processors_choices.face_editor_mouth_smile_range))
+		group_processors.add_argument('--face-editor-mouth-position-horizontal', help = wording.get('help.face_editor_mouth_position_horizontal'), type = float, default = config.get_float_value('processors', 'face_editor_mouth_position_horizontal', '0'), choices = processors_choices.face_editor_mouth_position_horizontal_range, metavar = create_float_metavar(processors_choices.face_editor_mouth_position_horizontal_range))
+		group_processors.add_argument('--face-editor-mouth-position-vertical', help = wording.get('help.face_editor_mouth_position_vertical'), type = float, default = config.get_float_value('processors', 'face_editor_mouth_position_vertical', '0'), choices = processors_choices.face_editor_mouth_position_vertical_range, metavar = create_float_metavar(processors_choices.face_editor_mouth_position_vertical_range))
+		group_processors.add_argument('--face-editor-head-pitch', help = wording.get('help.face_editor_head_pitch'), type = float, default = config.get_float_value('processors', 'face_editor_head_pitch', '0'), choices = processors_choices.face_editor_head_pitch_range, metavar = create_float_metavar(processors_choices.face_editor_head_pitch_range))
+		group_processors.add_argument('--face-editor-head-yaw', help = wording.get('help.face_editor_head_yaw'), type = float, default = config.get_float_value('processors', 'face_editor_head_yaw', '0'), choices = processors_choices.face_editor_head_yaw_range, metavar = create_float_metavar(processors_choices.face_editor_head_yaw_range))
+		group_processors.add_argument('--face-editor-head-roll', help = wording.get('help.face_editor_head_roll'), type = float, default = config.get_float_value('processors', 'face_editor_head_roll', '0'), choices = processors_choices.face_editor_head_roll_range, metavar = create_float_metavar(processors_choices.face_editor_head_roll_range))
+		facefusion.jobs.job_store.register_step_keys([ 'face_editor_model', 'face_editor_eyebrow_direction', 'face_editor_eye_gaze_horizontal', 'face_editor_eye_gaze_vertical', 'face_editor_eye_open_ratio', 'face_editor_lip_open_ratio', 'face_editor_mouth_grim', 'face_editor_mouth_pout', 'face_editor_mouth_purse', 'face_editor_mouth_smile', 'face_editor_mouth_position_horizontal', 'face_editor_mouth_position_vertical', 'face_editor_head_pitch', 'face_editor_head_yaw', 'face_editor_head_roll' ])
+
+
+def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
+	apply_state_item('face_editor_model', args.get('face_editor_model'))
+	apply_state_item('face_editor_eyebrow_direction', args.get('face_editor_eyebrow_direction'))
+	apply_state_item('face_editor_eye_gaze_horizontal', args.get('face_editor_eye_gaze_horizontal'))
+	apply_state_item('face_editor_eye_gaze_vertical', args.get('face_editor_eye_gaze_vertical'))
+	apply_state_item('face_editor_eye_open_ratio', args.get('face_editor_eye_open_ratio'))
+	apply_state_item('face_editor_lip_open_ratio', args.get('face_editor_lip_open_ratio'))
+	apply_state_item('face_editor_mouth_grim', args.get('face_editor_mouth_grim'))
+	apply_state_item('face_editor_mouth_pout', args.get('face_editor_mouth_pout'))
+	apply_state_item('face_editor_mouth_purse', args.get('face_editor_mouth_purse'))
+	apply_state_item('face_editor_mouth_smile', args.get('face_editor_mouth_smile'))
+	apply_state_item('face_editor_mouth_position_horizontal', args.get('face_editor_mouth_position_horizontal'))
+	apply_state_item('face_editor_mouth_position_vertical', args.get('face_editor_mouth_position_vertical'))
+	apply_state_item('face_editor_head_pitch', args.get('face_editor_head_pitch'))
+	apply_state_item('face_editor_head_yaw', args.get('face_editor_head_yaw'))
+	apply_state_item('face_editor_head_roll', args.get('face_editor_head_roll'))
+
+
+def pre_check() -> bool:
+	model_hash_set = get_model_options().get('hashes')
+	model_source_set = get_model_options().get('sources')
+
+	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)
+
+
+def pre_process(mode : ProcessMode) -> bool:
+	if mode in [ 'output', 'preview' ] and not is_image(state_manager.get_item('target_path')) and not is_video(state_manager.get_item('target_path')):
+		logger.error(wording.get('choose_image_or_video_target') + wording.get('exclamation_mark'), __name__)
+		return False
+	if mode == 'output' and not in_directory(state_manager.get_item('output_path')):
+		logger.error(wording.get('specify_image_or_video_output') + wording.get('exclamation_mark'), __name__)
+		return False
+	if mode == 'output' and not same_file_extension(state_manager.get_item('target_path'), state_manager.get_item('output_path')):
+		logger.error(wording.get('match_target_and_output_extension') + wording.get('exclamation_mark'), __name__)
+		return False
+	return True
+
+
+def post_process() -> None:
+	read_static_image.cache_clear()
+	video_manager.clear_video_pool()
+	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
+		clear_inference_pool()
+	if state_manager.get_item('video_memory_strategy') == 'strict':
+		content_analyser.clear_inference_pool()
+		face_classifier.clear_inference_pool()
+		face_detector.clear_inference_pool()
+		face_landmarker.clear_inference_pool()
+		face_masker.clear_inference_pool()
+		face_recognizer.clear_inference_pool()
+
+
+def edit_face(target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
+	model_template = get_model_options().get('template')
+	model_size = get_model_options().get('size')
+	face_landmark_5 = scale_face_landmark_5(target_face.landmark_set.get('5/68'), 1.5)
+	crop_vision_frame, affine_matrix = warp_face_by_face_landmark_5(temp_vision_frame, face_landmark_5, model_template, model_size)
+	box_mask = create_box_mask(crop_vision_frame, state_manager.get_item('face_mask_blur'), (0, 0, 0, 0))
+	crop_vision_frame = prepare_crop_frame(crop_vision_frame)
+	crop_vision_frame = apply_edit(crop_vision_frame, target_face.landmark_set.get('68'))
+	crop_vision_frame = normalize_crop_frame(crop_vision_frame)
+	temp_vision_frame = paste_back(temp_vision_frame, crop_vision_frame, box_mask, affine_matrix)
+	return temp_vision_frame
+
+
+def apply_edit(crop_vision_frame : VisionFrame, face_landmark_68 : FaceLandmark68) -> VisionFrame:
+	feature_volume = forward_extract_feature(crop_vision_frame)
+	pitch, yaw, roll, scale, translation, expression, motion_points = forward_extract_motion(crop_vision_frame)
+	rotation = create_rotation(pitch, yaw, roll)
+	motion_points_target = scale * (motion_points @ rotation.T + expression) + translation
+	expression = edit_eye_gaze(expression)
+	expression = edit_mouth_grim(expression)
+	expression = edit_mouth_position(expression)
+	expression = edit_mouth_pout(expression)
+	expression = edit_mouth_purse(expression)
+	expression = edit_mouth_smile(expression)
+	expression = edit_eyebrow_direction(expression)
+	expression = limit_expression(expression)
+	rotation = edit_head_rotation(pitch, yaw, roll)
+	motion_points_source = motion_points @ rotation.T
+	motion_points_source += expression
+	motion_points_source *= scale
+	motion_points_source += translation
+	motion_points_source += edit_eye_open(motion_points_target, face_landmark_68)
+	motion_points_source += edit_lip_open(motion_points_target, face_landmark_68)
+	motion_points_source = forward_stitch_motion_points(motion_points_source, motion_points_target)
+	crop_vision_frame = forward_generate_frame(feature_volume, motion_points_source, motion_points_target)
+	return crop_vision_frame
+
+
+def forward_extract_feature(crop_vision_frame : VisionFrame) -> LivePortraitFeatureVolume:
+	feature_extractor = get_inference_pool().get('feature_extractor')
+
+	with conditional_thread_semaphore():
+		feature_volume = feature_extractor.run(None,
+		{
+			'input': crop_vision_frame
+		})[0]
+
+	return feature_volume
+
+
+def forward_extract_motion(crop_vision_frame : VisionFrame) -> Tuple[LivePortraitPitch, LivePortraitYaw, LivePortraitRoll, LivePortraitScale, LivePortraitTranslation, LivePortraitExpression, LivePortraitMotionPoints]:
+	motion_extractor = get_inference_pool().get('motion_extractor')
+
+	with conditional_thread_semaphore():
+		pitch, yaw, roll, scale, translation, expression, motion_points = motion_extractor.run(None,
+		{
+			'input': crop_vision_frame
+		})
+
+	return pitch, yaw, roll, scale, translation, expression, motion_points
+
+
+def forward_retarget_eye(eye_motion_points : LivePortraitMotionPoints) -> LivePortraitMotionPoints:
+	eye_retargeter = get_inference_pool().get('eye_retargeter')
+
+	with conditional_thread_semaphore():
+		eye_motion_points = eye_retargeter.run(None,
+		{
+			'input': eye_motion_points
+		})[0]
+
+	return eye_motion_points
+
+
+def forward_retarget_lip(lip_motion_points : LivePortraitMotionPoints) -> LivePortraitMotionPoints:
+	lip_retargeter = get_inference_pool().get('lip_retargeter')
+
+	with conditional_thread_semaphore():
+		lip_motion_points = lip_retargeter.run(None,
+		{
+			'input': lip_motion_points
+		})[0]
+
+	return lip_motion_points
+
+
+def forward_stitch_motion_points(source_motion_points : LivePortraitMotionPoints, target_motion_points : LivePortraitMotionPoints) -> LivePortraitMotionPoints:
+	stitcher = get_inference_pool().get('stitcher')
+
+	with thread_semaphore():
+		motion_points = stitcher.run(None,
+		{
+			'source': source_motion_points,
+			'target': target_motion_points
+		})[0]
+
+	return motion_points
+
+
+def forward_generate_frame(feature_volume : LivePortraitFeatureVolume, source_motion_points : LivePortraitMotionPoints, target_motion_points : LivePortraitMotionPoints) -> VisionFrame:
+	generator = get_inference_pool().get('generator')
+
+	with thread_semaphore():
+		crop_vision_frame = generator.run(None,
+		{
+			'feature_volume': feature_volume,
+			'source': source_motion_points,
+			'target': target_motion_points
+		})[0][0]
+
+	return crop_vision_frame
+
+
+def edit_eyebrow_direction(expression : LivePortraitExpression) -> LivePortraitExpression:
+	face_editor_eyebrow = state_manager.get_item('face_editor_eyebrow_direction')
+
+	if face_editor_eyebrow > 0:
+		expression[0, 1, 1] += numpy.interp(face_editor_eyebrow, [ -1, 1 ], [ -0.015, 0.015 ])
+		expression[0, 2, 1] -= numpy.interp(face_editor_eyebrow, [ -1, 1 ], [ -0.020, 0.020 ])
+	else:
+		expression[0, 1, 0] -= numpy.interp(face_editor_eyebrow, [ -1, 1 ], [ -0.015, 0.015 ])
+		expression[0, 2, 0] += numpy.interp(face_editor_eyebrow, [ -1, 1 ], [ -0.020, 0.020 ])
+		expression[0, 1, 1] += numpy.interp(face_editor_eyebrow, [ -1, 1 ], [ -0.005, 0.005 ])
+		expression[0, 2, 1] -= numpy.interp(face_editor_eyebrow, [ -1, 1 ], [ -0.005, 0.005 ])
+	return expression
+
+
+def edit_eye_gaze(expression : LivePortraitExpression) -> LivePortraitExpression:
+	face_editor_eye_gaze_horizontal = state_manager.get_item('face_editor_eye_gaze_horizontal')
+	face_editor_eye_gaze_vertical = state_manager.get_item('face_editor_eye_gaze_vertical')
+
+	if face_editor_eye_gaze_horizontal > 0:
+		expression[0, 11, 0] += numpy.interp(face_editor_eye_gaze_horizontal, [ -1, 1 ], [ -0.015, 0.015 ])
+		expression[0, 15, 0] += numpy.interp(face_editor_eye_gaze_horizontal, [ -1, 1 ], [ -0.020, 0.020 ])
+	else:
+		expression[0, 11, 0] += numpy.interp(face_editor_eye_gaze_horizontal, [ -1, 1 ], [ -0.020, 0.020 ])
+		expression[0, 15, 0] += numpy.interp(face_editor_eye_gaze_horizontal, [ -1, 1 ], [ -0.015, 0.015 ])
+	expression[0, 1, 1] += numpy.interp(face_editor_eye_gaze_vertical, [ -1, 1 ], [ -0.0025, 0.0025 ])
+	expression[0, 2, 1] -= numpy.interp(face_editor_eye_gaze_vertical, [ -1, 1 ], [ -0.0025, 0.0025 ])
+	expression[0, 11, 1] -= numpy.interp(face_editor_eye_gaze_vertical, [ -1, 1 ], [ -0.010, 0.010 ])
+	expression[0, 13, 1] -= numpy.interp(face_editor_eye_gaze_vertical, [ -1, 1 ], [ -0.005, 0.005 ])
+	expression[0, 15, 1] -= numpy.interp(face_editor_eye_gaze_vertical, [ -1, 1 ], [ -0.010, 0.010 ])
+	expression[0, 16, 1] -= numpy.interp(face_editor_eye_gaze_vertical, [ -1, 1 ], [ -0.005, 0.005 ])
+	return expression
+
+
+def edit_eye_open(motion_points : LivePortraitMotionPoints, face_landmark_68 : FaceLandmark68) -> LivePortraitMotionPoints:
+	face_editor_eye_open_ratio = state_manager.get_item('face_editor_eye_open_ratio')
+	left_eye_ratio = calc_distance_ratio(face_landmark_68, 37, 40, 39, 36)
+	right_eye_ratio = calc_distance_ratio(face_landmark_68, 43, 46, 45, 42)
+
+	if face_editor_eye_open_ratio < 0:
+		eye_motion_points = numpy.concatenate([ motion_points.ravel(), [ left_eye_ratio, right_eye_ratio, 0.0 ] ])
+	else:
+		eye_motion_points = numpy.concatenate([ motion_points.ravel(), [ left_eye_ratio, right_eye_ratio, 0.6 ] ])
+	eye_motion_points = eye_motion_points.reshape(1, -1).astype(numpy.float32)
+	eye_motion_points = forward_retarget_eye(eye_motion_points) * numpy.abs(face_editor_eye_open_ratio)
+	eye_motion_points = eye_motion_points.reshape(-1, 21, 3)
+	return eye_motion_points
+
+
+def edit_lip_open(motion_points : LivePortraitMotionPoints, face_landmark_68 : FaceLandmark68) -> LivePortraitMotionPoints:
+	face_editor_lip_open_ratio = state_manager.get_item('face_editor_lip_open_ratio')
+	lip_ratio = calc_distance_ratio(face_landmark_68, 62, 66, 54, 48)
+
+	if face_editor_lip_open_ratio < 0:
+		lip_motion_points = numpy.concatenate([ motion_points.ravel(), [ lip_ratio, 0.0 ] ])
+	else:
+		lip_motion_points = numpy.concatenate([ motion_points.ravel(), [ lip_ratio, 1.0 ] ])
+	lip_motion_points = lip_motion_points.reshape(1, -1).astype(numpy.float32)
+	lip_motion_points = forward_retarget_lip(lip_motion_points) * numpy.abs(face_editor_lip_open_ratio)
+	lip_motion_points = lip_motion_points.reshape(-1, 21, 3)
+	return lip_motion_points
+
+
+def edit_mouth_grim(expression : LivePortraitExpression) -> LivePortraitExpression:
+	face_editor_mouth_grim = state_manager.get_item('face_editor_mouth_grim')
+	if face_editor_mouth_grim > 0:
+		expression[0, 17, 2] -= numpy.interp(face_editor_mouth_grim, [ -1, 1 ], [ -0.005, 0.005 ])
+		expression[0, 19, 2] += numpy.interp(face_editor_mouth_grim, [ -1, 1 ], [ -0.01, 0.01 ])
+		expression[0, 20, 1] -= numpy.interp(face_editor_mouth_grim, [ -1, 1 ], [ -0.06, 0.06 ])
+		expression[0, 20, 2] -= numpy.interp(face_editor_mouth_grim, [ -1, 1 ], [ -0.03, 0.03 ])
+	else:
+		expression[0, 19, 1] -= numpy.interp(face_editor_mouth_grim, [ -1, 1 ], [ -0.05, 0.05 ])
+		expression[0, 19, 2] -= numpy.interp(face_editor_mouth_grim, [ -1, 1 ], [ -0.02, 0.02 ])
+		expression[0, 20, 2] -= numpy.interp(face_editor_mouth_grim, [ -1, 1 ], [ -0.03, 0.03 ])
+	return expression
+
+
+def edit_mouth_position(expression : LivePortraitExpression) -> LivePortraitExpression:
+	face_editor_mouth_position_horizontal = state_manager.get_item('face_editor_mouth_position_horizontal')
+	face_editor_mouth_position_vertical = state_manager.get_item('face_editor_mouth_position_vertical')
+	expression[0, 19, 0] += numpy.interp(face_editor_mouth_position_horizontal, [ -1, 1 ], [ -0.05, 0.05 ])
+	expression[0, 20, 0] += numpy.interp(face_editor_mouth_position_horizontal, [ -1, 1 ], [ -0.04, 0.04 ])
+	if face_editor_mouth_position_vertical > 0:
+		expression[0, 19, 1] -= numpy.interp(face_editor_mouth_position_vertical, [ -1, 1 ], [ -0.04, 0.04 ])
+		expression[0, 20, 1] -= numpy.interp(face_editor_mouth_position_vertical, [ -1, 1 ], [ -0.02, 0.02 ])
+	else:
+		expression[0, 19, 1] -= numpy.interp(face_editor_mouth_position_vertical, [ -1, 1 ], [ -0.05, 0.05 ])
+		expression[0, 20, 1] -= numpy.interp(face_editor_mouth_position_vertical, [ -1, 1 ], [ -0.04, 0.04 ])
+	return expression
+
+
+def edit_mouth_pout(expression : LivePortraitExpression) -> LivePortraitExpression:
+	face_editor_mouth_pout = state_manager.get_item('face_editor_mouth_pout')
+	if face_editor_mouth_pout > 0:
+		expression[0, 19, 1] -= numpy.interp(face_editor_mouth_pout, [ -1, 1 ], [ -0.022, 0.022 ])
+		expression[0, 19, 2] += numpy.interp(face_editor_mouth_pout, [ -1, 1 ], [ -0.025, 0.025 ])
+		expression[0, 20, 2] -= numpy.interp(face_editor_mouth_pout, [ -1, 1 ], [ -0.002, 0.002 ])
+	else:
+		expression[0, 19, 1] += numpy.interp(face_editor_mouth_pout, [ -1, 1 ], [ -0.022, 0.022 ])
+		expression[0, 19, 2] += numpy.interp(face_editor_mouth_pout, [ -1, 1 ], [ -0.025, 0.025 ])
+		expression[0, 20, 2] -= numpy.interp(face_editor_mouth_pout, [ -1, 1 ], [ -0.002, 0.002 ])
+	return expression
+
+
+def edit_mouth_purse(expression : LivePortraitExpression) -> LivePortraitExpression:
+	face_editor_mouth_purse = state_manager.get_item('face_editor_mouth_purse')
+	if face_editor_mouth_purse > 0:
+		expression[0, 19, 1] -= numpy.interp(face_editor_mouth_purse, [ -1, 1 ], [ -0.04, 0.04 ])
+		expression[0, 19, 2] -= numpy.interp(face_editor_mouth_purse, [ -1, 1 ], [ -0.02, 0.02 ])
+	else:
+		expression[0, 14, 1] -= numpy.interp(face_editor_mouth_purse, [ -1, 1 ], [ -0.02, 0.02 ])
+		expression[0, 17, 2] += numpy.interp(face_editor_mouth_purse, [ -1, 1 ], [ -0.01, 0.01 ])
+		expression[0, 19, 2] -= numpy.interp(face_editor_mouth_purse, [ -1, 1 ], [ -0.015, 0.015 ])
+		expression[0, 20, 2] -= numpy.interp(face_editor_mouth_purse, [ -1, 1 ], [ -0.002, 0.002 ])
+	return expression
+
+
+def edit_mouth_smile(expression : LivePortraitExpression) -> LivePortraitExpression:
+	face_editor_mouth_smile = state_manager.get_item('face_editor_mouth_smile')
+	if face_editor_mouth_smile > 0:
+		expression[0, 20, 1] -= numpy.interp(face_editor_mouth_smile, [ -1, 1 ], [ -0.015, 0.015 ])
+		expression[0, 14, 1] -= numpy.interp(face_editor_mouth_smile, [ -1, 1 ], [ -0.025, 0.025 ])
+		expression[0, 17, 1] += numpy.interp(face_editor_mouth_smile, [ -1, 1 ], [ -0.01, 0.01 ])
+		expression[0, 17, 2] += numpy.interp(face_editor_mouth_smile, [ -1, 1 ], [ -0.004, 0.004 ])
+		expression[0, 3, 1] -= numpy.interp(face_editor_mouth_smile, [ -1, 1 ], [ -0.0045, 0.0045 ])
+		expression[0, 7, 1] -= numpy.interp(face_editor_mouth_smile, [ -1, 1 ], [ -0.0045, 0.0045 ])
+	else:
+		expression[0, 14, 1] -= numpy.interp(face_editor_mouth_smile, [ -1, 1 ], [ -0.02, 0.02 ])
+		expression[0, 17, 1] += numpy.interp(face_editor_mouth_smile, [ -1, 1 ], [ -0.003, 0.003 ])
+		expression[0, 19, 1] += numpy.interp(face_editor_mouth_smile, [ -1, 1 ], [ -0.02, 0.02 ])
+		expression[0, 19, 2] -= numpy.interp(face_editor_mouth_smile, [ -1, 1 ], [ -0.005, 0.005 ])
+		expression[0, 20, 2] += numpy.interp(face_editor_mouth_smile, [ -1, 1 ], [ -0.01, 0.01 ])
+		expression[0, 3, 1] += numpy.interp(face_editor_mouth_smile, [ -1, 1 ], [ -0.0045, 0.0045 ])
+		expression[0, 7, 1] += numpy.interp(face_editor_mouth_smile, [ -1, 1 ], [ -0.0045, 0.0045 ])
+	return expression
+
+
+def edit_head_rotation(pitch : LivePortraitPitch, yaw : LivePortraitYaw, roll : LivePortraitRoll) -> LivePortraitRotation:
+	face_editor_head_pitch = state_manager.get_item('face_editor_head_pitch')
+	face_editor_head_yaw = state_manager.get_item('face_editor_head_yaw')
+	face_editor_head_roll = state_manager.get_item('face_editor_head_roll')
+	edit_pitch = pitch + float(numpy.interp(face_editor_head_pitch, [ -1, 1 ], [ 20, -20 ]))
+	edit_yaw = yaw + float(numpy.interp(face_editor_head_yaw, [ -1, 1 ], [ 60, -60 ]))
+	edit_roll = roll + float(numpy.interp(face_editor_head_roll, [ -1, 1 ], [ -15, 15 ]))
+	edit_pitch, edit_yaw, edit_roll = limit_euler_angles(pitch, yaw, roll, edit_pitch, edit_yaw, edit_roll)
+	rotation = create_rotation(edit_pitch, edit_yaw, edit_roll)
+	return rotation
+
+
+def calc_distance_ratio(face_landmark_68 : FaceLandmark68, top_index : int, bottom_index : int, left_index : int, right_index : int) -> float:
+	vertical_direction = face_landmark_68[top_index] - face_landmark_68[bottom_index]
+	horizontal_direction = face_landmark_68[left_index] - face_landmark_68[right_index]
+	distance_ratio = float(numpy.linalg.norm(vertical_direction) / (numpy.linalg.norm(horizontal_direction) + 1e-6))
+	return distance_ratio
+
+
+def prepare_crop_frame(crop_vision_frame : VisionFrame) -> VisionFrame:
+	model_size = get_model_options().get('size')
+	prepare_size = (model_size[0] // 2, model_size[1] // 2)
+	crop_vision_frame = cv2.resize(crop_vision_frame, prepare_size, interpolation = cv2.INTER_AREA)
+	crop_vision_frame = crop_vision_frame[:, :, ::-1] / 255.0
+	crop_vision_frame = numpy.expand_dims(crop_vision_frame.transpose(2, 0, 1), axis = 0).astype(numpy.float32)
+	return crop_vision_frame
+
+
+def normalize_crop_frame(crop_vision_frame : VisionFrame) -> VisionFrame:
+	crop_vision_frame = crop_vision_frame.transpose(1, 2, 0).clip(0, 1)
+	crop_vision_frame = (crop_vision_frame * 255.0)
+	crop_vision_frame = crop_vision_frame.astype(numpy.uint8)[:, :, ::-1]
+	return crop_vision_frame
+
+
+def get_reference_frame(source_face : Face, target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
+	pass
+
+
+def process_frame(inputs : FaceEditorInputs) -> VisionFrame:
+	reference_faces = inputs.get('reference_faces')
+	target_vision_frame = inputs.get('target_vision_frame')
+	many_faces = sort_and_filter_faces(get_many_faces([ target_vision_frame ]))
+
+	if state_manager.get_item('face_selector_mode') == 'many':
+		if many_faces:
+			for target_face in many_faces:
+				target_vision_frame = edit_face(target_face, target_vision_frame)
+	if state_manager.get_item('face_selector_mode') == 'one':
+		target_face = get_one_face(many_faces)
+		if target_face:
+			target_vision_frame = edit_face(target_face, target_vision_frame)
+	if state_manager.get_item('face_selector_mode') == 'reference':
+		similar_faces = find_similar_faces(many_faces, reference_faces, state_manager.get_item('reference_face_distance'))
+		if similar_faces:
+			for similar_face in similar_faces:
+				target_vision_frame = edit_face(similar_face, target_vision_frame)
+	return target_vision_frame
+
+
+def process_frames(source_path : List[str], queue_payloads : List[QueuePayload], update_progress : UpdateProgress) -> None:
+	reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
+
+	for queue_payload in process_manager.manage(queue_payloads):
+		target_vision_path = queue_payload['frame_path']
+		target_vision_frame = read_image(target_vision_path)
+		output_vision_frame = process_frame(
+		{
+			'reference_faces': reference_faces,
+			'target_vision_frame': target_vision_frame
+		})
+		write_image(target_vision_path, output_vision_frame)
+		update_progress(1)
+
+
+def process_image(source_path : str, target_path : str, output_path : str) -> None:
+	reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
+	target_vision_frame = read_static_image(target_path)
+	output_vision_frame = process_frame(
+	{
+		'reference_faces': reference_faces,
+		'target_vision_frame': target_vision_frame
+	})
+	write_image(output_path, output_vision_frame)
+
+
+def process_video(source_paths : List[str], temp_frame_paths : List[str]) -> None:
+	processors.multi_process_frames(None, temp_frame_paths, process_frames)

facefusion/processors/modules/lip_syncer.py

@@ -0,0 +1,348 @@
+from argparse import ArgumentParser
+from functools import lru_cache
+from typing import List
+
+import cv2
+import numpy
+
+import facefusion.jobs.job_manager
+import facefusion.jobs.job_store
+import facefusion.processors.core as processors
+from facefusion import config, content_analyser, face_classifier, face_detector, face_landmarker, face_masker, face_recognizer, inference_manager, logger, process_manager, state_manager, video_manager, voice_extractor, wording
+from facefusion.audio import create_empty_audio_frame, get_voice_frame, read_static_voice
+from facefusion.common_helper import create_float_metavar
+from facefusion.common_helper import get_first
+from facefusion.download import conditional_download_hashes, conditional_download_sources, resolve_download_url
+from facefusion.face_analyser import get_many_faces, get_one_face
+from facefusion.face_helper import create_bounding_box, paste_back, warp_face_by_bounding_box, warp_face_by_face_landmark_5
+from facefusion.face_masker import create_area_mask, create_box_mask, create_occlusion_mask
+from facefusion.face_selector import find_similar_faces, sort_and_filter_faces
+from facefusion.face_store import get_reference_faces
+from facefusion.filesystem import filter_audio_paths, has_audio, in_directory, is_image, is_video, resolve_relative_path, same_file_extension
+from facefusion.processors import choices as processors_choices
+from facefusion.processors.types import LipSyncerInputs, LipSyncerWeight
+from facefusion.program_helper import find_argument_group
+from facefusion.thread_helper import conditional_thread_semaphore
+from facefusion.types import ApplyStateItem, Args, AudioFrame, BoundingBox, DownloadScope, Face, InferencePool, ModelOptions, ModelSet, ProcessMode, QueuePayload, UpdateProgress, VisionFrame
+from facefusion.vision import read_image, read_static_image, restrict_video_fps, write_image
+
+
+@lru_cache(maxsize = None)
+def create_static_model_set(download_scope : DownloadScope) -> ModelSet:
+	return\
+	{
+		'edtalk_256':
+		{
+			'hashes':
+			{
+				'lip_syncer':
+				{
+					'url': resolve_download_url('models-3.3.0', 'edtalk_256.hash'),
+					'path': resolve_relative_path('../.assets/models/edtalk_256.hash')
+				}
+			},
+			'sources':
+			{
+				'lip_syncer':
+				{
+					'url': resolve_download_url('models-3.3.0', 'edtalk_256.onnx'),
+					'path': resolve_relative_path('../.assets/models/edtalk_256.onnx')
+				}
+			},
+			'type': 'edtalk',
+			'size': (256, 256)
+		},
+		'wav2lip_96':
+		{
+			'hashes':
+			{
+				'lip_syncer':
+				{
+					'url': resolve_download_url('models-3.0.0', 'wav2lip_96.hash'),
+					'path': resolve_relative_path('../.assets/models/wav2lip_96.hash')
+				}
+			},
+			'sources':
+			{
+				'lip_syncer':
+				{
+					'url': resolve_download_url('models-3.0.0', 'wav2lip_96.onnx'),
+					'path': resolve_relative_path('../.assets/models/wav2lip_96.onnx')
+				}
+			},
+			'type': 'wav2lip',
+			'size': (96, 96)
+		},
+		'wav2lip_gan_96':
+		{
+			'hashes':
+			{
+				'lip_syncer':
+				{
+					'url': resolve_download_url('models-3.0.0', 'wav2lip_gan_96.hash'),
+					'path': resolve_relative_path('../.assets/models/wav2lip_gan_96.hash')
+				}
+			},
+			'sources':
+			{
+				'lip_syncer':
+				{
+					'url': resolve_download_url('models-3.0.0', 'wav2lip_gan_96.onnx'),
+					'path': resolve_relative_path('../.assets/models/wav2lip_gan_96.onnx')
+				}
+			},
+			'type': 'wav2lip',
+			'size': (96, 96)
+		}
+	}
+
+
+def get_inference_pool() -> InferencePool:
+	model_names = [ state_manager.get_item('lip_syncer_model') ]
+	model_source_set = get_model_options().get('sources')
+
+	return inference_manager.get_inference_pool(__name__, model_names, model_source_set)
+
+
+def clear_inference_pool() -> None:
+	model_names = [ state_manager.get_item('lip_syncer_model') ]
+	inference_manager.clear_inference_pool(__name__, model_names)
+
+
+def get_model_options() -> ModelOptions:
+	model_name = state_manager.get_item('lip_syncer_model')
+	return create_static_model_set('full').get(model_name)
+
+
+def register_args(program : ArgumentParser) -> None:
+	group_processors = find_argument_group(program, 'processors')
+	if group_processors:
+		group_processors.add_argument('--lip-syncer-model', help = wording.get('help.lip_syncer_model'), default = config.get_str_value('processors', 'lip_syncer_model', 'wav2lip_gan_96'), choices = processors_choices.lip_syncer_models)
+		group_processors.add_argument('--lip-syncer-weight', help = wording.get('help.lip_syncer_weight'), type = float, default = config.get_float_value('processors', 'lip_syncer_weight', '0.5'), choices = processors_choices.lip_syncer_weight_range, metavar = create_float_metavar(processors_choices.lip_syncer_weight_range))
+		facefusion.jobs.job_store.register_step_keys([ 'lip_syncer_model', 'lip_syncer_weight' ])
+
+
+def apply_args(args : Args, apply_state_item : ApplyStateItem) -> None:
+	apply_state_item('lip_syncer_model', args.get('lip_syncer_model'))
+	apply_state_item('lip_syncer_weight', args.get('lip_syncer_weight'))
+
+
+def pre_check() -> bool:
+	model_hash_set = get_model_options().get('hashes')
+	model_source_set = get_model_options().get('sources')
+
+	return conditional_download_hashes(model_hash_set) and conditional_download_sources(model_source_set)
+
+
+def pre_process(mode : ProcessMode) -> bool:
+	if not has_audio(state_manager.get_item('source_paths')):
+		logger.error(wording.get('choose_audio_source') + wording.get('exclamation_mark'), __name__)
+		return False
+	if mode in [ 'output', 'preview' ] and not is_image(state_manager.get_item('target_path')) and not is_video(state_manager.get_item('target_path')):
+		logger.error(wording.get('choose_image_or_video_target') + wording.get('exclamation_mark'), __name__)
+		return False
+	if mode == 'output' and not in_directory(state_manager.get_item('output_path')):
+		logger.error(wording.get('specify_image_or_video_output') + wording.get('exclamation_mark'), __name__)
+		return False
+	if mode == 'output' and not same_file_extension(state_manager.get_item('target_path'), state_manager.get_item('output_path')):
+		logger.error(wording.get('match_target_and_output_extension') + wording.get('exclamation_mark'), __name__)
+		return False
+	return True
+
+
+def post_process() -> None:
+	read_static_image.cache_clear()
+	read_static_voice.cache_clear()
+	video_manager.clear_video_pool()
+	if state_manager.get_item('video_memory_strategy') in [ 'strict', 'moderate' ]:
+		clear_inference_pool()
+	if state_manager.get_item('video_memory_strategy') == 'strict':
+		content_analyser.clear_inference_pool()
+		face_classifier.clear_inference_pool()
+		face_detector.clear_inference_pool()
+		face_landmarker.clear_inference_pool()
+		face_masker.clear_inference_pool()
+		face_recognizer.clear_inference_pool()
+		voice_extractor.clear_inference_pool()
+
+
+def sync_lip(target_face : Face, temp_audio_frame : AudioFrame, temp_vision_frame : VisionFrame) -> VisionFrame:
+	model_type = get_model_options().get('type')
+	model_size = get_model_options().get('size')
+	temp_audio_frame = prepare_audio_frame(temp_audio_frame)
+	crop_vision_frame, affine_matrix = warp_face_by_face_landmark_5(temp_vision_frame, target_face.landmark_set.get('5/68'), 'ffhq_512', (512, 512))
+	crop_masks = []
+
+	if 'occlusion' in state_manager.get_item('face_mask_types'):
+		occlusion_mask = create_occlusion_mask(crop_vision_frame)
+		crop_masks.append(occlusion_mask)
+
+	if model_type == 'edtalk':
+		lip_syncer_weight = numpy.array([ state_manager.get_item('lip_syncer_weight') ]).astype(numpy.float32)
+		box_mask = create_box_mask(crop_vision_frame, state_manager.get_item('face_mask_blur'), state_manager.get_item('face_mask_padding'))
+		crop_masks.append(box_mask)
+		crop_vision_frame = prepare_crop_frame(crop_vision_frame)
+		crop_vision_frame = forward_edtalk(temp_audio_frame, crop_vision_frame, lip_syncer_weight)
+		crop_vision_frame = normalize_crop_frame(crop_vision_frame)
+	if model_type == 'wav2lip':
+		face_landmark_68 = cv2.transform(target_face.landmark_set.get('68').reshape(1, -1, 2), affine_matrix).reshape(-1, 2)
+		area_mask = create_area_mask(crop_vision_frame, face_landmark_68, [ 'lower-face' ])
+		crop_masks.append(area_mask)
+		bounding_box = create_bounding_box(face_landmark_68)
+		bounding_box = resize_bounding_box(bounding_box, 1 / 8)
+		area_vision_frame, area_matrix = warp_face_by_bounding_box(crop_vision_frame, bounding_box, model_size)
+		area_vision_frame = prepare_crop_frame(area_vision_frame)
+		area_vision_frame = forward_wav2lip(temp_audio_frame, area_vision_frame)
+		area_vision_frame = normalize_crop_frame(area_vision_frame)
+		crop_vision_frame = cv2.warpAffine(area_vision_frame, cv2.invertAffineTransform(area_matrix), (512, 512), borderMode = cv2.BORDER_REPLICATE)
+
+	crop_mask = numpy.minimum.reduce(crop_masks)
+	paste_vision_frame = paste_back(temp_vision_frame, crop_vision_frame, crop_mask, affine_matrix)
+	return paste_vision_frame
+
+
+def forward_edtalk(temp_audio_frame : AudioFrame, crop_vision_frame : VisionFrame, lip_syncer_weight : LipSyncerWeight) -> VisionFrame:
+	lip_syncer = get_inference_pool().get('lip_syncer')
+
+	with conditional_thread_semaphore():
+		crop_vision_frame = lip_syncer.run(None,
+		{
+			'source': temp_audio_frame,
+			'target': crop_vision_frame,
+			'weight': lip_syncer_weight
+		})[0]
+
+	return crop_vision_frame
+
+
+def forward_wav2lip(temp_audio_frame : AudioFrame, area_vision_frame : VisionFrame) -> VisionFrame:
+	lip_syncer = get_inference_pool().get('lip_syncer')
+
+	with conditional_thread_semaphore():
+		area_vision_frame = lip_syncer.run(None,
+		{
+			'source': temp_audio_frame,
+			'target': area_vision_frame
+		})[0]
+
+	return area_vision_frame
+
+
+def prepare_audio_frame(temp_audio_frame : AudioFrame) -> AudioFrame:
+	model_type = get_model_options().get('type')
+	temp_audio_frame = numpy.maximum(numpy.exp(-5 * numpy.log(10)), temp_audio_frame)
+	temp_audio_frame = numpy.log10(temp_audio_frame) * 1.6 + 3.2
+	temp_audio_frame = temp_audio_frame.clip(-4, 4).astype(numpy.float32)
+
+	if model_type == 'wav2lip':
+		temp_audio_frame = temp_audio_frame * state_manager.get_item('lip_syncer_weight') * 2.0
+
+	temp_audio_frame = numpy.expand_dims(temp_audio_frame, axis = (0, 1))
+	return temp_audio_frame
+
+
+def prepare_crop_frame(crop_vision_frame : VisionFrame) -> VisionFrame:
+	model_type = get_model_options().get('type')
+	model_size = get_model_options().get('size')
+
+	if model_type == 'edtalk':
+		crop_vision_frame = cv2.resize(crop_vision_frame, model_size, interpolation = cv2.INTER_AREA)
+		crop_vision_frame = crop_vision_frame[:, :, ::-1] / 255.0
+		crop_vision_frame = numpy.expand_dims(crop_vision_frame.transpose(2, 0, 1), axis = 0).astype(numpy.float32)
+	if model_type == 'wav2lip':
+		crop_vision_frame = numpy.expand_dims(crop_vision_frame, axis = 0)
+		prepare_vision_frame = crop_vision_frame.copy()
+		prepare_vision_frame[:, model_size[0] // 2:] = 0
+		crop_vision_frame = numpy.concatenate((prepare_vision_frame, crop_vision_frame), axis = 3)
+		crop_vision_frame = crop_vision_frame.transpose(0, 3, 1, 2).astype('float32') / 255.0
+
+	return crop_vision_frame
+
+
+def resize_bounding_box(bounding_box : BoundingBox, aspect_ratio : float) -> BoundingBox:
+	x1, y1, x2, y2 = bounding_box
+	y1 -= numpy.abs(y2 - y1) * aspect_ratio
+	bounding_box[1] = max(y1, 0)
+	return bounding_box
+
+
+def normalize_crop_frame(crop_vision_frame : VisionFrame) -> VisionFrame:
+	model_type = get_model_options().get('type')
+	crop_vision_frame = crop_vision_frame[0].transpose(1, 2, 0)
+	crop_vision_frame = crop_vision_frame.clip(0, 1) * 255
+	crop_vision_frame = crop_vision_frame.astype(numpy.uint8)
+
+	if model_type == 'edtalk':
+		crop_vision_frame = crop_vision_frame[:, :, ::-1]
+		crop_vision_frame = cv2.resize(crop_vision_frame, (512, 512), interpolation = cv2.INTER_CUBIC)
+
+	return crop_vision_frame
+
+
+def get_reference_frame(source_face : Face, target_face : Face, temp_vision_frame : VisionFrame) -> VisionFrame:
+	pass
+
+
+def process_frame(inputs : LipSyncerInputs) -> VisionFrame:
+	reference_faces = inputs.get('reference_faces')
+	source_audio_frame = inputs.get('source_audio_frame')
+	target_vision_frame = inputs.get('target_vision_frame')
+	many_faces = sort_and_filter_faces(get_many_faces([ target_vision_frame ]))
+
+	if state_manager.get_item('face_selector_mode') == 'many':
+		if many_faces:
+			for target_face in many_faces:
+				target_vision_frame = sync_lip(target_face, source_audio_frame, target_vision_frame)
+	if state_manager.get_item('face_selector_mode') == 'one':
+		target_face = get_one_face(many_faces)
+		if target_face:
+			target_vision_frame = sync_lip(target_face, source_audio_frame, target_vision_frame)
+	if state_manager.get_item('face_selector_mode') == 'reference':
+		similar_faces = find_similar_faces(many_faces, reference_faces, state_manager.get_item('reference_face_distance'))
+		if similar_faces:
+			for similar_face in similar_faces:
+				target_vision_frame = sync_lip(similar_face, source_audio_frame, target_vision_frame)
+	return target_vision_frame
+
+
+def process_frames(source_paths : List[str], queue_payloads : List[QueuePayload], update_progress : UpdateProgress) -> None:
+	reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
+	source_audio_path = get_first(filter_audio_paths(source_paths))
+	temp_video_fps = restrict_video_fps(state_manager.get_item('target_path'), state_manager.get_item('output_video_fps'))
+
+	for queue_payload in process_manager.manage(queue_payloads):
+		frame_number = queue_payload.get('frame_number')
+		target_vision_path = queue_payload.get('frame_path')
+		source_audio_frame = get_voice_frame(source_audio_path, temp_video_fps, frame_number)
+		if not numpy.any(source_audio_frame):
+			source_audio_frame = create_empty_audio_frame()
+		target_vision_frame = read_image(target_vision_path)
+		output_vision_frame = process_frame(
+		{
+			'reference_faces': reference_faces,
+			'source_audio_frame': source_audio_frame,
+			'target_vision_frame': target_vision_frame
+		})
+		write_image(target_vision_path, output_vision_frame)
+		update_progress(1)
+
+
+def process_image(source_paths : List[str], target_path : str, output_path : str) -> None:
+	reference_faces = get_reference_faces() if 'reference' in state_manager.get_item('face_selector_mode') else None
+	source_audio_frame = create_empty_audio_frame()
+	target_vision_frame = read_static_image(target_path)
+	output_vision_frame = process_frame(
+	{
+		'reference_faces': reference_faces,
+		'source_audio_frame': source_audio_frame,
+		'target_vision_frame': target_vision_frame
+	})
+	write_image(output_path, output_vision_frame)
+
+
+def process_video(source_paths : List[str], temp_frame_paths : List[str]) -> None:
+	source_audio_paths = filter_audio_paths(state_manager.get_item('source_paths'))
+	temp_video_fps = restrict_video_fps(state_manager.get_item('target_path'), state_manager.get_item('output_video_fps'))
+	for source_audio_path in source_audio_paths:
+		read_static_voice(source_audio_path, temp_video_fps)
+	processors.multi_process_frames(source_paths, temp_frame_paths, process_frames)

facefusion/uis/layouts/default.py

@@ -1,7 +1,7 @@
 import gradio
 
 from facefusion import state_manager
-from facefusion.uis.components import about, common_options, download, face_detector, face_enhancer_options, face_masker, face_selector, face_swapper_options, frame_colorizer_options, frame_enhancer_options, instant_runner, job_manager, job_runner, output, output_options, preview, processors, source, target, terminal, trim_frame, ui_workflow #deep_swapper_options, age_modifier_options, memory, lip_syncer_options, face_debugger_options, temp_frame, face_landmarker, execution, execution_queue_count, execution_thread_count, expression_restorer_options, face_editor_options
+from facefusion.uis.components import about, age_modifier_options, common_options, deep_swapper_options, download, execution, execution_queue_count, execution_thread_count, expression_restorer_options, face_debugger_options, face_detector, face_editor_options, face_enhancer_options, face_landmarker, face_masker, face_selector, face_swapper_options, frame_colorizer_options, frame_enhancer_options, instant_runner, job_manager, job_runner, lip_syncer_options, memory, output, output_options, preview, processors, source, target, temp_frame, terminal, trim_frame, ui_workflow
 
 
 def pre_check() -> bool:
@@ -16,16 +16,16 @@ def render() -> gradio.Blocks:
 					about.render()
 				with gradio.Blocks():
 					processors.render()
-				#with gradio.Blocks():
-				#	age_modifier_options.render()
-				#with gradio.Blocks():
-				#	deep_swapper_options.render()
-				#with gradio.Blocks():
-				#	expression_restorer_options.render()
-				#with gradio.Blocks():
-				#	face_debugger_options.render()
-				#with gradio.Blocks():
-				#	face_editor_options.render()
+				with gradio.Blocks():
+					age_modifier_options.render()
+				with gradio.Blocks():
+					deep_swapper_options.render()
+				with gradio.Blocks():
+					expression_restorer_options.render()
+				with gradio.Blocks():
+					face_debugger_options.render()
+				with gradio.Blocks():
+					face_editor_options.render()
 				with gradio.Blocks():
 					face_enhancer_options.render()
 				with gradio.Blocks():
@@ -34,18 +34,18 @@ def render() -> gradio.Blocks:
 					frame_colorizer_options.render()
 				with gradio.Blocks():
 					frame_enhancer_options.render()
-				#with gradio.Blocks():
-				#	lip_syncer_options.render()
-				#with gradio.Blocks():
-				#	execution.render()
-				#	execution_thread_count.render()
-				#	execution_queue_count.render()
+				with gradio.Blocks():
+					lip_syncer_options.render()
+				with gradio.Blocks():
+					execution.render()
+					execution_thread_count.render()
+					execution_queue_count.render()
 				with gradio.Blocks():
 					download.render()
-				#with gradio.Blocks():
-				#	memory.render()
-				#with gradio.Blocks():
-				#	temp_frame.render()
+				with gradio.Blocks():
+					memory.render()
+				with gradio.Blocks():
+					temp_frame.render()
 				with gradio.Blocks():
 					output_options.render()
 			with gradio.Column(scale = 4):
@@ -57,11 +57,11 @@ def render() -> gradio.Blocks:
 					output.render()
 				with gradio.Blocks():
 					terminal.render()
-				#with gradio.Blocks():
-				#	ui_workflow.render()
-				#	instant_runner.render()
-				#	job_runner.render()
-				#	job_manager.render()
+				with gradio.Blocks():
+					ui_workflow.render()
+					instant_runner.render()
+					job_runner.render()
+					job_manager.render()
 			with gradio.Column(scale = 7):
 				with gradio.Blocks():
 					preview.render()
@@ -73,8 +73,8 @@ def render() -> gradio.Blocks:
 					face_masker.render()
 				with gradio.Blocks():
 					face_detector.render()
-				#with gradio.Blocks():
-				#	face_landmarker.render()
+				with gradio.Blocks():
+					face_landmarker.render()
 				with gradio.Blocks():
 					common_options.render()
 	return layout
@@ -82,36 +82,36 @@ def render() -> gradio.Blocks:
 
 def listen() -> None:
 	processors.listen()
-	#age_modifier_options.listen()
-	#deep_swapper_options.listen()
-	#expression_restorer_options.listen()
-	#face_debugger_options.listen()
-	#face_editor_options.listen()
+	age_modifier_options.listen()
+	deep_swapper_options.listen()
+	expression_restorer_options.listen()
+	face_debugger_options.listen()
+	face_editor_options.listen()
 	face_enhancer_options.listen()
 	face_swapper_options.listen()
 	frame_colorizer_options.listen()
 	frame_enhancer_options.listen()
-	#lip_syncer_options.listen()
-	#execution.listen()
-	#execution_thread_count.listen()
-	#execution_queue_count.listen()
+	lip_syncer_options.listen()
+	execution.listen()
+	execution_thread_count.listen()
+	execution_queue_count.listen()
 	download.listen()
-	#memory.listen()
-	#temp_frame.listen()
+	memory.listen()
+	temp_frame.listen()
 	output_options.listen()
 	source.listen()
 	target.listen()
 	output.listen()
-	#instant_runner.listen()
-	#job_runner.listen()
-	#job_manager.listen()
+	instant_runner.listen()
+	job_runner.listen()
+	job_manager.listen()
 	terminal.listen()
 	preview.listen()
 	trim_frame.listen()
 	face_selector.listen()
 	face_masker.listen()
 	face_detector.listen()
-	#face_landmarker.listen()
+	face_landmarker.listen()
 	common_options.listen()
 
 

facefusion/uis/types.py

@@ -3,15 +3,15 @@ from typing import Any, Dict, IO, Literal, TypeAlias
 File : TypeAlias = IO[Any]
 ComponentName = Literal\
 [
-	#'age_modifier_direction_slider',
-	#'age_modifier_model_dropdown',
-	#'benchmark_cycle_count_slider',
-	#'benchmark_resolutions_checkbox_group',
-	#'deep_swapper_model_dropdown',
-	#'deep_swapper_morph_slider',
-	#'expression_restorer_factor_slider',
-	#'expression_restorer_model_dropdown',
-	#'face_debugger_items_checkbox_group',
+	'age_modifier_direction_slider',
+	'age_modifier_model_dropdown',
+	'benchmark_cycle_count_slider',
+	'benchmark_resolutions_checkbox_group',
+	'deep_swapper_model_dropdown',
+	'deep_swapper_morph_slider',
+	'expression_restorer_factor_slider',
+	'expression_restorer_model_dropdown',
+	'face_debugger_items_checkbox_group',
 	'face_detector_angles_checkbox_group',
 	'face_detector_model_dropdown',
 	'face_detector_score_slider',
@@ -34,8 +34,8 @@ ComponentName = Literal\
 	'face_enhancer_blend_slider',
 	'face_enhancer_model_dropdown',
 	'face_enhancer_weight_slider',
-	#'face_landmarker_model_dropdown',
-	#'face_landmarker_score_slider',
+	'face_landmarker_model_dropdown',
+	'face_landmarker_score_slider',
 	'face_mask_types_checkbox_group',
 	'face_mask_areas_checkbox_group',
 	'face_mask_regions_checkbox_group',
@@ -73,10 +73,10 @@ ComponentName = Literal\
 	'target_image',
 	'target_video',
 	'ui_workflow_dropdown',
-	#'webcam_device_id_dropdown',
-	#'webcam_fps_slider',
-	#'webcam_mode_radio',
-	#'webcam_resolution_dropdown'
+	'webcam_device_id_dropdown',
+	'webcam_fps_slider',
+	'webcam_mode_radio',
+	'webcam_resolution_dropdown'
 ]
 Component : TypeAlias = Any
 ComponentOptions : TypeAlias = Dict[str, Any]