Spaces:
Running
Running
import gradio as gr | |
from gradio_toggle import Toggle | |
import torch | |
from huggingface_hub import snapshot_download | |
from transformers import pipeline | |
from xora.models.autoencoders.causal_video_autoencoder import CausalVideoAutoencoder | |
from xora.models.transformers.transformer3d import Transformer3DModel | |
from xora.models.transformers.symmetric_patchifier import SymmetricPatchifier | |
from xora.schedulers.rf import RectifiedFlowScheduler | |
from xora.pipelines.pipeline_xora_video import XoraVideoPipeline | |
from transformers import T5EncoderModel, T5Tokenizer | |
from xora.utils.conditioning_method import ConditioningMethod | |
from pathlib import Path | |
import safetensors.torch | |
import json | |
import numpy as np | |
import cv2 | |
from PIL import Image | |
import tempfile | |
import os | |
import gc | |
from openai import OpenAI | |
import re | |
# Load system prompts | |
system_prompt_t2v = """๋น์ ์ ๋น๋์ค ์์ฑ์ ์ํ ํ๋กฌํํธ ์ ๋ฌธ๊ฐ์ ๋๋ค. | |
์ฃผ์ด์ง ํ๋กฌํํธ๋ฅผ ๋ค์ ๊ตฌ์กฐ์ ๋ง๊ฒ ๊ฐ์ ํด์ฃผ์ธ์: | |
1. ์ฃผ์ ๋์์ ๋ช ํํ ํ ๋ฌธ์ฅ์ผ๋ก ์์ | |
2. ๊ตฌ์ฒด์ ์ธ ๋์๊ณผ ์ ์ค์ฒ๋ฅผ ์๊ฐ ์์๋๋ก ์ค๋ช | |
3. ์บ๋ฆญํฐ/๊ฐ์ฒด์ ์ธ๋ชจ๋ฅผ ์์ธํ ๋ฌ์ฌ | |
4. ๋ฐฐ๊ฒฝ๊ณผ ํ๊ฒฝ ์ธ๋ถ ์ฌํญ์ ๊ตฌ์ฒด์ ์ผ๋ก ํฌํจ | |
5. ์นด๋ฉ๋ผ ๊ฐ๋์ ์์ง์์ ๋ช ์ | |
6. ์กฐ๋ช ๊ณผ ์์์ ์์ธํ ์ค๋ช | |
7. ๋ณํ๋ ๊ฐ์์ค๋ฌ์ด ์ฌ๊ฑด์ ์์ฐ์ค๋ฝ๊ฒ ํฌํจ | |
๋ชจ๋ ์ค๋ช ์ ํ๋์ ์์ฐ์ค๋ฌ์ด ๋ฌธ๋จ์ผ๋ก ์์ฑํ๊ณ , | |
์ดฌ์ ๊ฐ๋ ์ด ์ดฌ์ ๋ชฉ๋ก์ ์ค๋ช ํ๋ ๊ฒ์ฒ๋ผ ๊ตฌ์ฒด์ ์ด๊ณ ์๊ฐ์ ์ผ๋ก ์์ฑํ์ธ์. | |
200๋จ์ด๋ฅผ ๋์ง ์๋๋ก ํ๋, ์ต๋ํ ์์ธํ๊ฒ ์์ฑํ์ธ์.""" | |
system_prompt_i2v = """๋น์ ์ ์ด๋ฏธ์ง ๊ธฐ๋ฐ ๋น๋์ค ์์ฑ์ ์ํ ํ๋กฌํํธ ์ ๋ฌธ๊ฐ์ ๋๋ค. | |
์ฃผ์ด์ง ํ๋กฌํํธ๋ฅผ ๋ค์ ๊ตฌ์กฐ์ ๋ง๊ฒ ๊ฐ์ ํด์ฃผ์ธ์: | |
1. ์ฃผ์ ๋์์ ๋ช ํํ ํ ๋ฌธ์ฅ์ผ๋ก ์์ | |
2. ๊ตฌ์ฒด์ ์ธ ๋์๊ณผ ์ ์ค์ฒ๋ฅผ ์๊ฐ ์์๋๋ก ์ค๋ช | |
3. ์บ๋ฆญํฐ/๊ฐ์ฒด์ ์ธ๋ชจ๋ฅผ ์์ธํ ๋ฌ์ฌ | |
4. ๋ฐฐ๊ฒฝ๊ณผ ํ๊ฒฝ ์ธ๋ถ ์ฌํญ์ ๊ตฌ์ฒด์ ์ผ๋ก ํฌํจ | |
5. ์นด๋ฉ๋ผ ๊ฐ๋์ ์์ง์์ ๋ช ์ | |
6. ์กฐ๋ช ๊ณผ ์์์ ์์ธํ ์ค๋ช | |
7. ๋ณํ๋ ๊ฐ์์ค๋ฌ์ด ์ฌ๊ฑด์ ์์ฐ์ค๋ฝ๊ฒ ํฌํจ | |
๋ชจ๋ ์ค๋ช ์ ํ๋์ ์์ฐ์ค๋ฌ์ด ๋ฌธ๋จ์ผ๋ก ์์ฑํ๊ณ , | |
์ดฌ์ ๊ฐ๋ ์ด ์ดฌ์ ๋ชฉ๋ก์ ์ค๋ช ํ๋ ๊ฒ์ฒ๋ผ ๊ตฌ์ฒด์ ์ด๊ณ ์๊ฐ์ ์ผ๋ก ์์ฑํ์ธ์. | |
200๋จ์ด๋ฅผ ๋์ง ์๋๋ก ํ๋, ์ต๋ํ ์์ธํ๊ฒ ์์ฑํ์ธ์.""" | |
# Load Hugging Face token if needed | |
hf_token = os.getenv("HF_TOKEN") | |
openai_api_key = os.getenv("OPENAI_API_KEY") | |
client = OpenAI(api_key=openai_api_key) | |
# Initialize translation pipeline with device and clean_up settings | |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu") | |
translator = pipeline( | |
"translation", | |
model="Helsinki-NLP/opus-mt-ko-en", | |
device=device, | |
clean_up_tokenization_spaces=True | |
) | |
# Korean text detection function | |
def contains_korean(text): | |
korean_pattern = re.compile('[ใฑ-ใ ใ -ใ ฃ๊ฐ-ํฃ]') | |
return bool(korean_pattern.search(text)) | |
def translate_korean_prompt(prompt, max_length=450): | |
""" | |
Translate Korean prompt to English if Korean text is detected | |
Split long text into chunks if necessary | |
""" | |
if not contains_korean(prompt): | |
return prompt | |
# Split long text into chunks | |
def split_text(text, max_length): | |
words = text.split() | |
chunks = [] | |
current_chunk = [] | |
current_length = 0 | |
for word in words: | |
if current_length + len(word) + 1 > max_length: | |
chunks.append(' '.join(current_chunk)) | |
current_chunk = [word] | |
current_length = len(word) | |
else: | |
current_chunk.append(word) | |
current_length += len(word) + 1 | |
if current_chunk: | |
chunks.append(' '.join(current_chunk)) | |
return chunks | |
try: | |
if len(prompt) > max_length: | |
chunks = split_text(prompt, max_length) | |
translated_chunks = [] | |
for chunk in chunks: | |
translated = translator(chunk, max_length=512)[0]['translation_text'] | |
translated_chunks.append(translated) | |
final_translation = ' '.join(translated_chunks) | |
else: | |
final_translation = translator(prompt, max_length=512)[0]['translation_text'] | |
print(f"Original Korean prompt: {prompt}") | |
print(f"Translated English prompt: {final_translation}") | |
return final_translation | |
except Exception as e: | |
print(f"Translation error: {e}") | |
return prompt # Return original prompt if translation fails | |
def enhance_prompt(prompt, type="t2v"): | |
system_prompt = system_prompt_t2v if type == "t2v" else system_prompt_i2v | |
messages = [ | |
{"role": "system", "content": system_prompt}, | |
{"role": "user", "content": prompt}, | |
] | |
try: | |
response = client.chat.completions.create( | |
model="gpt-4-1106-preview", | |
messages=messages, | |
max_tokens=2000, | |
) | |
enhanced_prompt = response.choices[0].message.content.strip() | |
print("\n=== ํ๋กฌํํธ ์ฆ๊ฐ ๊ฒฐ๊ณผ ===") | |
print("Original Prompt:") | |
print(prompt) | |
print("\nEnhanced Prompt:") | |
print(enhanced_prompt) | |
print("========================\n") | |
return enhanced_prompt | |
except Exception as e: | |
print(f"Error during prompt enhancement: {e}") | |
return prompt | |
def update_prompt_t2v(prompt, enhance_toggle): | |
return update_prompt(prompt, enhance_toggle, "t2v") | |
def update_prompt_i2v(prompt, enhance_toggle): | |
return update_prompt(prompt, enhance_toggle, "i2v") | |
def update_prompt(prompt, enhance_toggle, type="t2v"): | |
if enhance_toggle: | |
return enhance_prompt(prompt, type) | |
return prompt | |
# Set model download directory within Hugging Face Spaces | |
model_path = "asset" | |
if not os.path.exists(model_path): | |
snapshot_download( | |
"Lightricks/LTX-Video", local_dir=model_path, repo_type="model", token=hf_token | |
) | |
# Global variables to load components | |
vae_dir = Path(model_path) / "vae" | |
unet_dir = Path(model_path) / "unet" | |
scheduler_dir = Path(model_path) / "scheduler" | |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu") | |
def load_vae(vae_dir): | |
vae_ckpt_path = vae_dir / "vae_diffusion_pytorch_model.safetensors" | |
vae_config_path = vae_dir / "config.json" | |
with open(vae_config_path, "r") as f: | |
vae_config = json.load(f) | |
vae = CausalVideoAutoencoder.from_config(vae_config) | |
vae_state_dict = safetensors.torch.load_file(vae_ckpt_path) | |
vae.load_state_dict(vae_state_dict) | |
return vae.to(device=device, dtype=torch.bfloat16) | |
def load_unet(unet_dir): | |
unet_ckpt_path = unet_dir / "unet_diffusion_pytorch_model.safetensors" | |
unet_config_path = unet_dir / "config.json" | |
transformer_config = Transformer3DModel.load_config(unet_config_path) | |
transformer = Transformer3DModel.from_config(transformer_config) | |
unet_state_dict = safetensors.torch.load_file(unet_ckpt_path) | |
transformer.load_state_dict(unet_state_dict, strict=True) | |
return transformer.to(device=device, dtype=torch.bfloat16) | |
def load_scheduler(scheduler_dir): | |
scheduler_config_path = scheduler_dir / "scheduler_config.json" | |
scheduler_config = RectifiedFlowScheduler.load_config(scheduler_config_path) | |
return RectifiedFlowScheduler.from_config(scheduler_config) | |
# Helper function for image processing | |
def center_crop_and_resize(frame, target_height, target_width): | |
h, w, _ = frame.shape | |
aspect_ratio_target = target_width / target_height | |
aspect_ratio_frame = w / h | |
if aspect_ratio_frame > aspect_ratio_target: | |
new_width = int(h * aspect_ratio_target) | |
x_start = (w - new_width) // 2 | |
frame_cropped = frame[:, x_start : x_start + new_width] | |
else: | |
new_height = int(w / aspect_ratio_target) | |
y_start = (h - new_height) // 2 | |
frame_cropped = frame[y_start : y_start + new_height, :] | |
frame_resized = cv2.resize(frame_cropped, (target_width, target_height)) | |
return frame_resized | |
def load_image_to_tensor_with_resize(image_path, target_height=512, target_width=768): | |
image = Image.open(image_path).convert("RGB") | |
image_np = np.array(image) | |
frame_resized = center_crop_and_resize(image_np, target_height, target_width) | |
frame_tensor = torch.tensor(frame_resized).permute(2, 0, 1).float() | |
frame_tensor = (frame_tensor / 127.5) - 1.0 | |
return frame_tensor.unsqueeze(0).unsqueeze(2) | |
# Load models | |
vae = load_vae(vae_dir) | |
unet = load_unet(unet_dir) | |
scheduler = load_scheduler(scheduler_dir) | |
patchifier = SymmetricPatchifier(patch_size=1) | |
text_encoder = T5EncoderModel.from_pretrained( | |
"PixArt-alpha/PixArt-XL-2-1024-MS", subfolder="text_encoder" | |
).to(device) | |
tokenizer = T5Tokenizer.from_pretrained( | |
"PixArt-alpha/PixArt-XL-2-1024-MS", subfolder="tokenizer" | |
) | |
pipeline = XoraVideoPipeline( | |
transformer=unet, | |
patchifier=patchifier, | |
text_encoder=text_encoder, | |
tokenizer=tokenizer, | |
scheduler=scheduler, | |
vae=vae, | |
).to(device) | |
# State ๋ณ์๋ค์ ์ด๊ธฐํ ์์ | |
txt2vid_current_height = gr.State(value=320) | |
txt2vid_current_width = gr.State(value=512) | |
txt2vid_current_num_frames = gr.State(value=257) | |
img2vid_current_height = gr.State(value=320) | |
img2vid_current_width = gr.State(value=512) | |
img2vid_current_num_frames = gr.State(value=257) | |
# Preset options for resolution and frame configuration | |
# Convert frames to seconds assuming 25 FPS | |
preset_options = [ | |
{"label": "[16:9 HD] 1216x704, 1.6์ด", "width": 1216, "height": 704, "num_frames": 41}, | |
{"label": "[16:9] 1088x704, 2.0์ด", "width": 1088, "height": 704, "num_frames": 49}, | |
{"label": "[16:9] 1056x640, 2.3์ด", "width": 1056, "height": 640, "num_frames": 57}, | |
{"label": "[16:9] 992x608, 2.6์ด", "width": 992, "height": 608, "num_frames": 65}, | |
{"label": "[16:9] 896x608, 2.9์ด", "width": 896, "height": 608, "num_frames": 73}, | |
{"label": "[16:9] 896x544, 3.2์ด", "width": 896, "height": 544, "num_frames": 81}, | |
{"label": "[16:9] 832x544, 3.6์ด", "width": 832, "height": 544, "num_frames": 89}, | |
{"label": "[16:9] 800x512, 3.9์ด", "width": 800, "height": 512, "num_frames": 97}, | |
{"label": "[16:9] 768x512, 3.9์ด", "width": 768, "height": 512, "num_frames": 97}, | |
{"label": "[16:9] 800x480, 4.2์ด", "width": 800, "height": 480, "num_frames": 105}, | |
{"label": "[16:9] 736x480, 4.5์ด", "width": 736, "height": 480, "num_frames": 113}, | |
{"label": "[3:2] 704x480, 4.8์ด", "width": 704, "height": 480, "num_frames": 121}, | |
{"label": "[16:9] 704x448, 5.2์ด", "width": 704, "height": 448, "num_frames": 129}, | |
{"label": "[16:9] 672x448, 5.5์ด", "width": 672, "height": 448, "num_frames": 137}, | |
{"label": "[16:9] 640x416, 6.1์ด", "width": 640, "height": 416, "num_frames": 153}, | |
{"label": "[16:9] 672x384, 6.4์ด", "width": 672, "height": 384, "num_frames": 161}, | |
{"label": "[16:9] 640x384, 6.8์ด", "width": 640, "height": 384, "num_frames": 169}, | |
{"label": "[16:9] 608x384, 7.1์ด", "width": 608, "height": 384, "num_frames": 177}, | |
{"label": "[16:9] 576x384, 7.4์ด", "width": 576, "height": 384, "num_frames": 185}, | |
{"label": "[16:9] 608x352, 7.7์ด", "width": 608, "height": 352, "num_frames": 193}, | |
{"label": "[16:9] 576x352, 8.0์ด", "width": 576, "height": 352, "num_frames": 201}, | |
{"label": "[16:9] 544x352, 8.4์ด", "width": 544, "height": 352, "num_frames": 209}, | |
{"label": "[3:2] 512x352, 9.3์ด", "width": 512, "height": 352, "num_frames": 233}, | |
{"label": "[16:9] 544x320, 9.6์ด", "width": 544, "height": 320, "num_frames": 241}, | |
{"label": "[16:9] 512x320, 10.3์ด", "width": 512, "height": 320, "num_frames": 257}, | |
] | |
def preset_changed(preset): | |
selected = next(item for item in preset_options if item["label"] == preset) | |
return [ | |
selected["height"], | |
selected["width"], | |
selected["num_frames"], | |
gr.update(visible=False), | |
gr.update(visible=False), | |
gr.update(visible=False), | |
] | |
def generate_video_from_text( | |
prompt="", | |
enhance_prompt_toggle=False, | |
negative_prompt="low quality, worst quality, deformed, distorted, warped, motion smear, motion artifacts, fused fingers, incorrect anatomy, strange hands, unattractive", | |
frame_rate=25, | |
seed=171198, | |
num_inference_steps=41, | |
guidance_scale=4, | |
height=320, | |
width=512, | |
num_frames=257, | |
progress=gr.Progress(), | |
): | |
if len(prompt.strip()) < 50: | |
raise gr.Error( | |
"ํ๋กฌํํธ๋ ์ต์ 50์ ์ด์์ด์ด์ผ ํฉ๋๋ค. ๋ ์์ธํ ์ค๋ช ์ ์ ๊ณตํด์ฃผ์ธ์.", | |
duration=5, | |
) | |
# Translate Korean prompts to English | |
prompt = translate_korean_prompt(prompt) | |
negative_prompt = translate_korean_prompt(negative_prompt) | |
sample = { | |
"prompt": prompt, | |
"prompt_attention_mask": None, | |
"negative_prompt": negative_prompt, | |
"negative_prompt_attention_mask": None, | |
"media_items": None, | |
} | |
generator = torch.Generator(device="cpu").manual_seed(seed) | |
def gradio_progress_callback(self, step, timestep, kwargs): | |
progress((step + 1) / num_inference_steps) | |
try: | |
with torch.no_grad(): | |
images = pipeline( | |
num_inference_steps=num_inference_steps, | |
num_images_per_prompt=1, | |
guidance_scale=guidance_scale, | |
generator=generator, | |
output_type="pt", | |
height=height, | |
width=width, | |
num_frames=num_frames, | |
frame_rate=frame_rate, | |
**sample, | |
is_video=True, | |
vae_per_channel_normalize=True, | |
conditioning_method=ConditioningMethod.UNCONDITIONAL, | |
mixed_precision=True, | |
callback_on_step_end=gradio_progress_callback, | |
).images | |
except Exception as e: | |
raise gr.Error( | |
f"๋น๋์ค ์์ฑ ์ค ์ค๋ฅ๊ฐ ๋ฐ์ํ์ต๋๋ค. ๋ค์ ์๋ํด์ฃผ์ธ์. ์ค๋ฅ: {e}", | |
duration=5, | |
) | |
finally: | |
torch.cuda.empty_cache() | |
gc.collect() | |
output_path = tempfile.mktemp(suffix=".mp4") | |
print(images.shape) | |
video_np = images.squeeze(0).permute(1, 2, 3, 0).cpu().float().numpy() | |
video_np = (video_np * 255).astype(np.uint8) | |
height, width = video_np.shape[1:3] | |
out = cv2.VideoWriter( | |
output_path, cv2.VideoWriter_fourcc(*"mp4v"), frame_rate, (width, height) | |
) | |
for frame in video_np[..., ::-1]: | |
out.write(frame) | |
out.release() | |
del images | |
del video_np | |
torch.cuda.empty_cache() | |
return output_path | |
def generate_video_from_image( | |
image_path, | |
prompt="", | |
enhance_prompt_toggle=False, | |
negative_prompt="low quality, worst quality, deformed, distorted, warped, motion smear, motion artifacts, fused fingers, incorrect anatomy, strange hands, unattractive", | |
frame_rate=25, | |
seed=171198, | |
num_inference_steps=41, | |
guidance_scale=4, | |
height=320, | |
width=512, | |
num_frames=257, | |
progress=gr.Progress(), | |
): | |
print("Height: ", height) | |
print("Width: ", width) | |
print("Num Frames: ", num_frames) | |
if len(prompt.strip()) < 50: | |
raise gr.Error( | |
"ํ๋กฌํํธ๋ ์ต์ 50์ ์ด์์ด์ด์ผ ํฉ๋๋ค. ๋ ์์ธํ ์ค๋ช ์ ์ ๊ณตํด์ฃผ์ธ์.", | |
duration=5, | |
) | |
if not image_path: | |
raise gr.Error("์ ๋ ฅ ์ด๋ฏธ์ง๋ฅผ ์ ๊ณตํด์ฃผ์ธ์.", duration=5) | |
# Translate Korean prompts to English | |
prompt = translate_korean_prompt(prompt) | |
negative_prompt = translate_korean_prompt(negative_prompt) | |
media_items = ( | |
load_image_to_tensor_with_resize(image_path, height, width).to(device).detach() | |
) | |
sample = { | |
"prompt": prompt, | |
"prompt_attention_mask": None, | |
"negative_prompt": negative_prompt, | |
"negative_prompt_attention_mask": None, | |
"media_items": media_items, | |
} | |
generator = torch.Generator(device="cpu").manual_seed(seed) | |
def gradio_progress_callback(self, step, timestep, kwargs): | |
progress((step + 1) / num_inference_steps) | |
try: | |
with torch.no_grad(): | |
images = pipeline( | |
num_inference_steps=num_inference_steps, | |
num_images_per_prompt=1, | |
guidance_scale=guidance_scale, | |
generator=generator, | |
output_type="pt", | |
height=height, | |
width=width, | |
num_frames=num_frames, | |
frame_rate=frame_rate, | |
**sample, | |
is_video=True, | |
vae_per_channel_normalize=True, | |
conditioning_method=ConditioningMethod.FIRST_FRAME, | |
mixed_precision=True, | |
callback_on_step_end=gradio_progress_callback, | |
).images | |
output_path = tempfile.mktemp(suffix=".mp4") | |
video_np = images.squeeze(0).permute(1, 2, 3, 0).cpu().float().numpy() | |
video_np = (video_np * 255).astype(np.uint8) | |
height, width = video_np.shape[1:3] | |
out = cv2.VideoWriter( | |
output_path, cv2.VideoWriter_fourcc(*"mp4v"), frame_rate, (width, height) | |
) | |
for frame in video_np[..., ::-1]: | |
out.write(frame) | |
out.release() | |
except Exception as e: | |
raise gr.Error( | |
f"๋น๋์ค ์์ฑ ์ค ์ค๋ฅ๊ฐ ๋ฐ์ํ์ต๋๋ค. ๋ค์ ์๋ํด์ฃผ์ธ์. ์ค๋ฅ: {e}", | |
duration=5, | |
) | |
finally: | |
torch.cuda.empty_cache() | |
gc.collect() | |
return output_path | |
def create_advanced_options(): | |
with gr.Accordion("Step 4: Advanced Options (Optional)", open=False): | |
seed = gr.Slider( | |
label="Seed", | |
minimum=0, | |
maximum=1000000, | |
step=1, | |
value=171198 | |
) | |
inference_steps = gr.Slider( | |
label="4.2 Inference Steps", | |
minimum=1, | |
maximum=50, | |
step=1, | |
value=41, | |
visible=False | |
) | |
guidance_scale = gr.Slider( | |
label="4.3 Guidance Scale", | |
minimum=1.0, | |
maximum=5.0, | |
step=0.1, | |
value=4.0, | |
visible=False | |
) | |
height_slider = gr.Slider( | |
label="4.4 Height", | |
minimum=256, | |
maximum=1024, | |
step=64, | |
value=320, | |
visible=False, | |
) | |
width_slider = gr.Slider( | |
label="4.5 Width", | |
minimum=256, | |
maximum=1024, | |
step=64, | |
value=512, | |
visible=False, | |
) | |
num_frames_slider = gr.Slider( | |
label="4.5 Number of Frames", | |
minimum=1, | |
maximum=200, | |
step=1, | |
value=257, | |
visible=False, | |
) | |
return [ | |
seed, | |
inference_steps, | |
guidance_scale, | |
height_slider, | |
width_slider, | |
num_frames_slider, | |
] | |
system_prompt_scenario = """๋น์ ์ ์์ ์คํฌ๋ฆฝํธ์ ๋ง๋ ๋ฐฐ๊ฒฝ ์์์ ์์ฑํ๊ธฐ ์ํ ํ๋กฌํํธ ์ ๋ฌธ๊ฐ์ ๋๋ค. | |
์ฃผ์ด์ง ์คํฌ๋ฆฝํธ์ ๋ถ์๊ธฐ์ ๋งฅ๋ฝ์ ์๊ฐ์ ๋ฐฐ๊ฒฝ์ผ๋ก ํํํ๋, ๋ค์ ์์น์ ๋ฐ๋์ ์ค์ํ์ธ์: | |
1. ์ ํ์ด๋ ์๋น์ค๋ฅผ ์ง์ ์ ์ผ๋ก ๋ฌ์ฌํ์ง ๋ง ๊ฒ | |
2. ์คํฌ๋ฆฝํธ์ ๊ฐ์ฑ๊ณผ ํค์ค๋งค๋๋ฅผ ํํํ๋ ๋ฐฐ๊ฒฝ ์์์ ์ง์คํ ๊ฒ | |
3. 5๊ฐ ์น์ ์ด ํ๋์ ์ด์ผ๊ธฐ์ฒ๋ผ ์์ฐ์ค๋ฝ๊ฒ ์ฐ๊ฒฐ๋๋๋ก ํ ๊ฒ | |
4. ์ถ์์ ์ด๊ณ ์์ ์ ์ธ ์๊ฐ ํํ์ ํ์ฉํ ๊ฒ | |
๊ฐ ์น์ ๋ณ ํ๋กฌํํธ ์์ฑ ๊ฐ์ด๋: | |
1. ๋ฐฐ๊ฒฝ ๋ฐ ํ์์ฑ: ์ฃผ์ ์ ์ ๋ฐ์ ์ธ ๋ถ์๊ธฐ๋ฅผ ํํํ๋ ๋ฐฐ๊ฒฝ ์ฌ | |
2. ๋ฌธ์ ์ ๊ธฐ: ๊ธด์ฅ๊ฐ์ด๋ ๊ฐ๋ฑ์ ์์ํ๋ ๋ถ์๊ธฐ ์๋ ๋ฐฐ๊ฒฝ | |
3. ํด๊ฒฐ์ฑ ์ ์: ํฌ๋ง์ ์ด๊ณ ๋ฐ์ ํค์ ๋ฐฐ๊ฒฝ ์ ํ | |
4. ๋ณธ๋ก : ์์ ๊ฐ ์๊ณ ์ ๋ขฐ๋๋ฅผ ๋์ด๋ ๋ฐฐ๊ฒฝ | |
5. ๊ฒฐ๋ก : ์ํฉํธ ์๋ ๋ง๋ฌด๋ฆฌ๋ฅผ ์ํ ์ญ๋์ ์ธ ๋ฐฐ๊ฒฝ | |
๋ชจ๋ ์น์ ์ด ์ผ๊ด๋ ์คํ์ผ๊ณผ ํค์ ์ ์งํ๋ฉด์๋ ์์ฐ์ค๋ฝ๊ฒ ์ด์ด์ง๋๋ก ๊ตฌ์ฑํ์ธ์. | |
๊ฐ ์น์ ์ ํ๋กฌํํธ ์์ฑ์ ๋ฐ๋์ ๋ค์ ๊ตฌ์กฐ์ ๋ง๊ฒ ๊ฐ์ ํด์ฃผ์ธ์: | |
1. ์ฃผ์ ๋์์ ๋ช ํํ ํ ๋ฌธ์ฅ์ผ๋ก ์์ | |
2. ๊ตฌ์ฒด์ ์ธ ๋์๊ณผ ์ ์ค์ฒ๋ฅผ ์๊ฐ ์์๋๋ก ์ค๋ช | |
3. ์บ๋ฆญํฐ/๊ฐ์ฒด์ ์ธ๋ชจ๋ฅผ ์์ธํ ๋ฌ์ฌ | |
4. ๋ฐฐ๊ฒฝ๊ณผ ํ๊ฒฝ ์ธ๋ถ ์ฌํญ์ ๊ตฌ์ฒด์ ์ผ๋ก ํฌํจ | |
5. ์นด๋ฉ๋ผ ๊ฐ๋์ ์์ง์์ ๋ช ์ | |
6. ์กฐ๋ช ๊ณผ ์์์ ์์ธํ ์ค๋ช | |
7. ๋ณํ๋ ๊ฐ์์ค๋ฌ์ด ์ฌ๊ฑด์ ์์ฐ์ค๋ฝ๊ฒ ํฌํจ | |
๋ชจ๋ ์ค๋ช ์ ํ๋์ ์์ฐ์ค๋ฌ์ด ๋ฌธ๋จ์ผ๋ก ์์ฑํ๊ณ , | |
์ดฌ์ ๊ฐ๋ ์ด ์ดฌ์ ๋ชฉ๋ก์ ์ค๋ช ํ๋ ๊ฒ์ฒ๋ผ ๊ตฌ์ฒด์ ์ด๊ณ ์๊ฐ์ ์ผ๋ก ์์ฑํ์ธ์. | |
200๋จ์ด๋ฅผ ๋์ง ์๋๋ก ํ๋, ์ต๋ํ ์์ธํ๊ฒ ์์ฑํ์ธ์. | |
""" | |
def analyze_scenario(scenario): | |
"""์๋๋ฆฌ์ค๋ฅผ ๋ถ์ํ์ฌ ๋ฐฐ๊ฒฝ ์์์ฉ ํ๋กฌํํธ ์์ฑ""" | |
messages = [ | |
{"role": "system", "content": system_prompt_scenario}, | |
{"role": "user", "content": f""" | |
๋ค์ ์คํฌ๋ฆฝํธ์ ๋ถ์๊ธฐ์ ๊ฐ์ฑ์ ํํํ ์ ์๋ ๋ฐฐ๊ฒฝ ์์ ํ๋กฌํํธ๋ฅผ ์์ฑํด์ฃผ์ธ์: | |
์ฃผ์ด์ง ์คํฌ๋ฆฝํธ์ ๋ถ์๊ธฐ์ ๋งฅ๋ฝ์ ์๊ฐ์ ๋ฐฐ๊ฒฝ์ผ๋ก ํํํ๋, ๋ค์ ์์น์ ๋ฐ๋์ ์ค์ํ์ธ์: | |
1. ์ ํ์ด๋ ์๋น์ค๋ฅผ ์ง์ ์ ์ผ๋ก ๋ฌ์ฌํ์ง ๋ง ๊ฒ | |
2. ์คํฌ๋ฆฝํธ์ ๊ฐ์ฑ๊ณผ ํค์ค๋งค๋๋ฅผ ํํํ๋ ๋ฐฐ๊ฒฝ ์์์ ์ง์คํ ๊ฒ | |
3. 5๊ฐ ์น์ ์ด ํ๋์ ์ด์ผ๊ธฐ์ฒ๋ผ ์์ฐ์ค๋ฝ๊ฒ ์ฐ๊ฒฐ๋๋๋ก ํ ๊ฒ | |
4. ์ถ์์ ์ด๊ณ ์์ ์ ์ธ ์๊ฐ ํํ์ ํ์ฉํ ๊ฒ | |
๊ฐ ์น์ ๋ณ ํ๋กฌํํธ ์์ฑ ๊ฐ์ด๋: | |
1. ๋ฐฐ๊ฒฝ ๋ฐ ํ์์ฑ: ์ฃผ์ ์ ์ ๋ฐ์ ์ธ ๋ถ์๊ธฐ๋ฅผ ํํํ๋ ๋ฐฐ๊ฒฝ ์ฌ | |
2. ํฅ๋ฏธ ์ ๋ฐ: ๊ธด์ฅ๊ฐ์ด๋ ๊ฐ๋ฑ์ ์์ํ๋ ๋ถ์๊ธฐ ์๋ ๋ฐฐ๊ฒฝ | |
3. ํด๊ฒฐ์ฑ ์ ์: ํฌ๋ง์ ์ด๊ณ ๋ฐ์ ํค์ ๋ฐฐ๊ฒฝ ์ ํ | |
4. ๋ณธ๋ก : ์์ ๊ฐ ์๊ณ ์ ๋ขฐ๋๋ฅผ ๋์ด๋ ๋ฐฐ๊ฒฝ | |
5. ๊ฒฐ๋ก : ์ํฉํธ ์๋ ๋ง๋ฌด๋ฆฌ๋ฅผ ์ํ ์ญ๋์ ์ธ ๋ฐฐ๊ฒฝ | |
๋ชจ๋ ์น์ ์ด ์ผ๊ด๋ ์คํ์ผ๊ณผ ํค์ ์ ์งํ๋ฉด์๋ ์์ฐ์ค๋ฝ๊ฒ ์ด์ด์ง๋๋ก ๊ตฌ์ฑํ์ธ์. | |
๊ฐ ์น์ ์ ํ๋กฌํํธ ์์ฑ์ ๋ฐ๋์ ๋ค์ ๊ตฌ์กฐ์ ๋ง๊ฒ ๊ฐ์ ํด์ฃผ์ธ์: | |
1. ์ฃผ์ ๋์์ ๋ช ํํ ํ ๋ฌธ์ฅ์ผ๋ก ์์ | |
2. ๊ตฌ์ฒด์ ์ธ ๋์๊ณผ ์ ์ค์ฒ๋ฅผ ์๊ฐ ์์๋๋ก ์ค๋ช | |
3. ์บ๋ฆญํฐ/๊ฐ์ฒด์ ์ธ๋ชจ๋ฅผ ์์ธํ ๋ฌ์ฌ | |
4. ๋ฐฐ๊ฒฝ๊ณผ ํ๊ฒฝ ์ธ๋ถ ์ฌํญ์ ๊ตฌ์ฒด์ ์ผ๋ก ํฌํจ | |
5. ์นด๋ฉ๋ผ ๊ฐ๋์ ์์ง์์ ๋ช ์ | |
6. ์กฐ๋ช ๊ณผ ์์์ ์์ธํ ์ค๋ช | |
7. ๋ณํ๋ ๊ฐ์์ค๋ฌ์ด ์ฌ๊ฑด์ ์์ฐ์ค๋ฝ๊ฒ ํฌํจ | |
๋ชจ๋ ์ค๋ช ์ ํ๋์ ์์ฐ์ค๋ฌ์ด ๋ฌธ๋จ์ผ๋ก ์์ฑํ๊ณ , | |
์ดฌ์ ๊ฐ๋ ์ด ์ดฌ์ ๋ชฉ๋ก์ ์ค๋ช ํ๋ ๊ฒ์ฒ๋ผ ๊ตฌ์ฒด์ ์ด๊ณ ์๊ฐ์ ์ผ๋ก ์์ฑํ์ธ์. | |
200๋จ์ด๋ฅผ ๋์ง ์๋๋ก ํ๋, ์ต๋ํ ์์ธํ๊ฒ ์์ฑํ์ธ์. | |
{scenario} | |
๊ฐ ์น์ ๋ณ๋ก ์ง์ ์ ์ธ ์ ํ ๋ฌ์ฌ๋ ํผํ๊ณ , ์คํฌ๋ฆฝํธ์ ๊ฐ์ฑ์ ํํํ๋ ๋ฐฐ๊ฒฝ ์์์ ์ง์คํด์ฃผ์ธ์."""}, | |
] | |
response = client.chat.completions.create( | |
model="gpt-4-1106-preview", | |
messages=messages, | |
max_tokens=2000, | |
) | |
content = response.choices[0].message.content.strip() | |
# ์น์ ๋ณ๋ก ๋ถ๋ฆฌ | |
sections = [] | |
current_section = "" | |
for line in content.split('\n'): | |
if line.strip().startswith(('1.', '2.', '3.', '4.', '5.')): | |
if current_section: | |
sections.append(current_section.strip()) | |
current_section = line | |
else: | |
current_section += "\n" + line | |
if current_section: | |
sections.append(current_section.strip()) | |
# ๋ถ์กฑํ ์น์ ์ฑ์ฐ๊ธฐ | |
while len(sections) < 5: | |
sections.append("์ถ๊ฐ ์น์ ์ด ํ์ํฉ๋๋ค.") | |
return sections[:5] | |
except Exception as e: | |
print(f"Error during scenario analysis: {e}") | |
return ["Error occurred during analysis"] * 5 | |
def generate_section_video(prompt, preset, section_number=1, base_seed=171198, progress=gr.Progress()): | |
"""๊ฐ ์น์ ์ ๋น๋์ค ์์ฑ - ์๋ฌ ์ฒ๋ฆฌ ์ถ๊ฐ""" | |
try: | |
if not prompt or len(prompt.strip()) < 50: | |
raise gr.Error("ํ๋กฌํํธ๋ ์ต์ 50์ ์ด์์ด์ด์ผ ํฉ๋๋ค.") | |
selected = next(item for item in preset_options if item["label"] == preset) | |
section_seed = base_seed + section_number | |
return generate_video_from_text( | |
prompt=prompt, | |
height=selected["height"], | |
width=selected["width"], | |
num_frames=selected["num_frames"], | |
seed=section_seed, | |
progress=progress | |
) | |
except Exception as e: | |
print(f"Error in section {section_number}: {e}") | |
raise gr.Error(f"์น์ {section_number} ์์ฑ ์ค ์ค๋ฅ: {str(e)}") | |
# ๊ฐ๋ณ ์น์ ํ๋กฌํํธ ์์ฑ ํจ์ ์ถ๊ฐ | |
def generate_single_section_prompt(scenario, section_number): | |
"""๊ฐ๋ณ ์น์ ์ ๋ํ ํ๋กฌํํธ ์์ฑ""" | |
section_descriptions = { | |
1: "๋ฐฐ๊ฒฝ ๋ฐ ํ์์ฑ: ์ฃผ์ ์ ์ ๋ฐ์ ์ธ ๋ถ์๊ธฐ๋ฅผ ํํํ๋ ๋ฐฐ๊ฒฝ ์ฌ", | |
2: "ํฅ๋ฏธ ์ ๋ฐ: ํฅ๋ฏธ๋ฅผ ์ ๋ฐํ๊ณ ๊ธฐ๋๊ฐ์ ์ฆํญ์ํค๋ ๋ฐฐ๊ฒฝ", | |
3: "ํด๊ฒฐ์ฑ ์ ์: ํฌ๋ง์ ์ด๊ณ ๋ฐ์ ํค์ ๋ฐฐ๊ฒฝ ์ ํ", | |
4: "๋ณธ๋ก : ์์ ๊ฐ ์๊ณ ์ ๋ขฐ๋๋ฅผ ๋์ด๋ ๋ฐฐ๊ฒฝ", | |
5: "๊ฒฐ๋ก : ์ํฉํธ ์๋ ๋ง๋ฌด๋ฆฌ๋ฅผ ์ํ ์ญ๋์ ์ธ ๋ฐฐ๊ฒฝ" | |
} | |
messages = [ | |
{"role": "system", "content": system_prompt_scenario}, | |
{"role": "user", "content": f""" | |
๋ค์ ์คํฌ๋ฆฝํธ์ {section_number}๋ฒ์งธ ์น์ ({section_descriptions[section_number]})์ ๋ํ | |
๋ฐฐ๊ฒฝ ์์ ํ๋กฌํํธ๋ง์ ์์ฑํด์ฃผ์ธ์: | |
{scenario} | |
์ง์ ์ ์ธ ์ ํ ๋ฌ์ฌ๋ ํผํ๊ณ , ์คํฌ๋ฆฝํธ์ ์ฃผ์ ์ ๊ฐ์ฑ์ ํํํ๋ ํต์ฌ ํค์๋๋ฅผ ๋ฐ์ํ ๋ฐฐ๊ฒฝ ์์์ ์ง์คํด์ฃผ์ธ์."""} | |
] | |
try: | |
response = client.chat.completions.create( | |
model="gpt-4-1106-preview", | |
messages=messages, | |
max_tokens=500, | |
) | |
return response.choices[0].message.content.strip() | |
except Exception as e: | |
print(f"Error during prompt generation: {e}") | |
return "Error occurred during prompt generation" | |
# ๋น๋์ค ๊ฒฐํฉ ํจ์ ์ถ๊ฐ | |
def combine_videos(video_paths, output_path): | |
"""์ฌ๋ฌ ๋น๋์ค๋ฅผ ํ๋๋ก ๊ฒฐํฉ""" | |
if not all(video_paths): | |
raise gr.Error("๋ชจ๋ ์น์ ์ ์์์ด ์์ฑ๋์ด์ผ ํฉ๋๋ค.") | |
try: | |
# ์ฒซ ๋ฒ์งธ ๋น๋์ค์ ์์ฑ ๊ฐ์ ธ์ค๊ธฐ | |
cap = cv2.VideoCapture(video_paths[0]) | |
fps = int(cap.get(cv2.CAP_PROP_FPS)) | |
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)) | |
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) | |
cap.release() | |
# ์ถ๋ ฅ ๋น๋์ค ์ค์ | |
fourcc = cv2.VideoWriter_fourcc(*'mp4v') | |
out = cv2.VideoWriter(output_path, fourcc, fps, (width, height)) | |
# ๊ฐ ๋น๋์ค ์์ฐจ์ ์ผ๋ก ๊ฒฐํฉ | |
for video_path in video_paths: | |
if video_path and os.path.exists(video_path): | |
cap = cv2.VideoCapture(video_path) | |
while True: | |
ret, frame = cap.read() | |
if not ret: | |
break | |
out.write(frame) | |
cap.release() | |
out.release() | |
return output_path | |
except Exception as e: | |
raise gr.Error(f"๋น๋์ค ๊ฒฐํฉ ์ค ์ค๋ฅ ๋ฐ์: {e}") | |
def merge_section_videos(section1, section2, section3, section4, section5): | |
"""์น์ ๋น๋์ค๋ค์ ํ๋๋ก ๊ฒฐํฉ""" | |
videos = [] | |
# ๊ฐ ์น์ ๋น๋์ค ํ์ธ ๋ฐ ์ฒ๋ฆฌ | |
for i, video in enumerate([section1, section2, section3, section4, section5], 1): | |
if video and os.path.exists(video): | |
videos.append(video) | |
else: | |
raise gr.Error(f"์น์ {i}์ ์์์ด ์๊ฑฐ๋ ์ ๊ทผํ ์ ์์ต๋๋ค.") | |
if not videos: | |
raise gr.Error("๊ฒฐํฉํ ์์์ด ์์ต๋๋ค.") | |
try: | |
output_path = tempfile.mktemp(suffix=".mp4") | |
# ์ฒซ ๋ฒ์งธ ๋น๋์ค์ ์์ฑ ๊ฐ์ ธ์ค๊ธฐ | |
cap = cv2.VideoCapture(videos[0]) | |
fps = int(cap.get(cv2.CAP_PROP_FPS)) | |
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)) | |
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) | |
cap.release() | |
# ์ถ๋ ฅ ๋น๋์ค ์ค์ | |
fourcc = cv2.VideoWriter_fourcc(*'mp4v') | |
out = cv2.VideoWriter(output_path, fourcc, fps, (width, height)) | |
# ๊ฐ ๋น๋์ค ์์ฐจ์ ์ผ๋ก ๊ฒฐํฉ | |
for video_path in videos: | |
cap = cv2.VideoCapture(video_path) | |
while True: | |
ret, frame = cap.read() | |
if not ret: | |
break | |
# ํ์ํ ๊ฒฝ์ฐ ํ๋ ์ ํฌ๊ธฐ ์กฐ์ | |
if frame.shape[:2] != (height, width): | |
frame = cv2.resize(frame, (width, height)) | |
out.write(frame) | |
cap.release() | |
out.release() | |
return output_path | |
except Exception as e: | |
raise gr.Error(f"๋น๋์ค ๊ฒฐํฉ ์ค ์ค๋ฅ ๋ฐ์: {e}") | |
# Gradio Interface Definition | |
with gr.Blocks(theme=gr.themes.Soft()) as iface: | |
with gr.Tabs(): | |
# Text to Video Tab | |
with gr.TabItem("ํ ์คํธ๋ก ๋น๋์ค ๋ง๋ค๊ธฐ"): | |
with gr.Row(): | |
with gr.Column(): | |
txt2vid_prompt = gr.Textbox( | |
label="Step 1: ํ๋กฌํํธ ์ ๋ ฅ", | |
placeholder="์์ฑํ๊ณ ์ถ์ ๋น๋์ค๋ฅผ ์ค๋ช ํ์ธ์ (์ต์ 50์)...", | |
value="๊ท์ฌ์ด ๊ณ ์์ด", | |
lines=5, | |
) | |
txt2vid_enhance_toggle = Toggle( | |
label="ํ๋กฌํํธ ๊ฐ์ ", | |
value=False, | |
interactive=True, | |
) | |
txt2vid_negative_prompt = gr.Textbox( | |
label="Step 2: ๋ค๊ฑฐํฐ๋ธ ํ๋กฌํํธ ์ ๋ ฅ", | |
placeholder="๋น๋์ค์์ ์ํ์ง ์๋ ์์๋ฅผ ์ค๋ช ํ์ธ์...", | |
value="low quality, worst quality, deformed, distorted, warped, motion smear, motion artifacts, fused fingers, incorrect anatomy, strange hands, unattractive", | |
lines=2, | |
visible=False | |
) | |
txt2vid_preset = gr.Dropdown( | |
choices=[p["label"] for p in preset_options], | |
value="[16:9] 512x320, 10.3์ด", | |
label="Step 2: ํด์๋ ํ๋ฆฌ์ ์ ํ", | |
) | |
txt2vid_frame_rate = gr.Slider( | |
label="Step 3: ํ๋ ์ ๋ ์ดํธ", | |
minimum=21, | |
maximum=30, | |
step=1, | |
value=25, | |
visible=False | |
) | |
txt2vid_advanced = create_advanced_options() | |
txt2vid_generate = gr.Button( | |
"Step 3: ๋น๋์ค ์์ฑ", | |
variant="primary", | |
size="lg", | |
) | |
with gr.Column(): | |
txt2vid_output = gr.Video(label="์์ฑ๋ ๋น๋์ค") | |
# Image to Video Tab | |
with gr.TabItem("์ด๋ฏธ์ง๋ก ๋น๋์ค ๋ง๋ค๊ธฐ"): | |
with gr.Row(): | |
with gr.Column(): | |
img2vid_image = gr.Image( | |
type="filepath", | |
label="Step 1: ์ ๋ ฅ ์ด๋ฏธ์ง ์ ๋ก๋", | |
elem_id="image_upload", | |
) | |
img2vid_prompt = gr.Textbox( | |
label="Step 2: ํ๋กฌํํธ ์ ๋ ฅ", | |
placeholder="์ด๋ฏธ์ง๋ฅผ ์ด๋ป๊ฒ ์ ๋๋ฉ์ด์ ํํ ์ง ์ค๋ช ํ์ธ์ (์ต์ 50์)...", | |
value="๊ท์ฌ์ด ๊ณ ์์ด", | |
lines=5, | |
) | |
img2vid_enhance_toggle = Toggle( | |
label="ํ๋กฌํํธ ์ฆ๊ฐ", | |
value=False, | |
interactive=True, | |
) | |
img2vid_negative_prompt = gr.Textbox( | |
label="Step 3: ๋ค๊ฑฐํฐ๋ธ ํ๋กฌํํธ ์ ๋ ฅ", | |
placeholder="๋น๋์ค์์ ์ํ์ง ์๋ ์์๋ฅผ ์ค๋ช ํ์ธ์...", | |
value="low quality, worst quality, deformed, distorted, warped, motion smear, motion artifacts, fused fingers, incorrect anatomy, strange hands, unattractive", | |
lines=2, | |
visible=False | |
) | |
img2vid_preset = gr.Dropdown( | |
choices=[p["label"] for p in preset_options], | |
value="[16:9] 512x320, 10.3์ด", | |
label="Step 3: ํด์๋ ํ๋ฆฌ์ ์ ํ", | |
) | |
img2vid_frame_rate = gr.Slider( | |
label="Step 4: ํ๋ ์ ๋ ์ดํธ", | |
minimum=21, | |
maximum=30, | |
step=1, | |
value=25, | |
visible=False | |
) | |
img2vid_advanced = create_advanced_options() | |
img2vid_generate = gr.Button( | |
"Step 4: ๋น๋์ค ์์ฑ", | |
variant="primary", | |
size="lg", | |
) | |
with gr.Column(): | |
img2vid_output = gr.Video(label="์์ฑ๋ ๋น๋์ค") | |
# Scenario to Video Tab (Modified) | |
with gr.TabItem("์๋๋ฆฌ์ค๋ก ๋น๋์ค ๋ง๋ค๊ธฐ(์ํผ)"): | |
with gr.Row(): | |
with gr.Column(scale=1): | |
scenario_input = gr.Textbox( | |
label="์์ ์คํฌ๋ฆฝํธ ์ ๋ ฅ", | |
placeholder="์ ์ฒด ์๋๋ฆฌ์ค๋ฅผ ์ ๋ ฅํ์ธ์...", | |
lines=10 | |
) | |
scenario_preset = gr.Dropdown( | |
choices=[p["label"] for p in preset_options], | |
value="[16:9] 512x320, 10.3์ด", | |
label="ํ๋ฉด ํฌ๊ธฐ ์ ํ" | |
) | |
analyze_btn = gr.Button("์๋๋ฆฌ์ค ๋ถ์ ๋ฐ ํ๋กฌํํธ ์์ฑ", variant="primary") | |
with gr.Column(scale=2): | |
with gr.Row(): | |
# ์น์ 1 | |
with gr.Column(): | |
section1_prompt = gr.Textbox( | |
label="1. ๋ฐฐ๊ฒฝ ๋ฐ ํ์์ฑ", | |
lines=4 | |
) | |
with gr.Row(): | |
section1_regenerate = gr.Button("๐ ํ๋กฌํํธ ์์ฑ") | |
section1_generate = gr.Button("๐ ์์ ์์ฑ") | |
section1_video = gr.Video(label="์น์ 1 ์์") | |
# ์น์ 2 | |
with gr.Column(): | |
section2_prompt = gr.Textbox( | |
label="2. ํฅ๋ฏธ ์ ๋ฐ", | |
lines=4 | |
) | |
with gr.Row(): | |
section2_regenerate = gr.Button("๐ ํ๋กฌํํธ ์์ฑ") | |
section2_generate = gr.Button("๐ ์์ ์์ฑ") | |
section2_video = gr.Video(label="์น์ 2 ์์") | |
with gr.Row(): | |
# ์น์ 3 | |
with gr.Column(): | |
section3_prompt = gr.Textbox( | |
label="3. ํด๊ฒฐ์ฑ ์ ์", | |
lines=4 | |
) | |
with gr.Row(): | |
section3_regenerate = gr.Button("๐ ํ๋กฌํํธ ์์ฑ") | |
section3_generate = gr.Button("๐ ์์ ์์ฑ") | |
section3_video = gr.Video(label="์น์ 3 ์์") | |
# ์น์ 4 | |
with gr.Column(): | |
section4_prompt = gr.Textbox( | |
label="4. ๋ณธ๋ก ", | |
lines=4 | |
) | |
with gr.Row(): | |
section4_regenerate = gr.Button("๐ ํ๋กฌํํธ ์์ฑ") | |
section4_generate = gr.Button("๐ ์์ ์์ฑ") | |
section4_video = gr.Video(label="์น์ 4 ์์") | |
with gr.Row(): | |
# ์น์ 5 | |
with gr.Column(): | |
section5_prompt = gr.Textbox( | |
label="5. ๊ฒฐ๋ก ๋ฐ ๊ฐ์กฐ", | |
lines=4 | |
) | |
with gr.Row(): | |
section5_regenerate = gr.Button("๐ ํ๋กฌํํธ ์์ฑ") | |
section5_generate = gr.Button("๐ ์์ ์์ฑ") | |
section5_video = gr.Video(label="์น์ 5 ์์") | |
# ํตํฉ ์์ ์น์ ์ถ๊ฐ | |
with gr.Row(): | |
with gr.Column(scale=1): | |
# ๊ธฐ์กด์ scenario_input๊ณผ analyze_btn ์ ์ง | |
merge_videos_btn = gr.Button("ํตํฉ ์์ ์์ฑ", variant="primary", size="lg") | |
with gr.Column(scale=2): | |
# ๊ธฐ์กด์ ์น์ 1-5 ์ ์ง | |
# ํตํฉ ์์ ์ถ๋ ฅ ์น์ ์ถ๊ฐ | |
with gr.Row(): | |
merged_video_output = gr.Video(label="ํตํฉ ์์") | |
# Event handlers | |
txt2vid_preset.change( | |
fn=preset_changed, | |
inputs=[txt2vid_preset], | |
outputs=[ | |
txt2vid_current_height, | |
txt2vid_current_width, | |
txt2vid_current_num_frames, | |
*txt2vid_advanced[3:] | |
] | |
) | |
txt2vid_enhance_toggle.change( | |
fn=update_prompt_t2v, | |
inputs=[txt2vid_prompt, txt2vid_enhance_toggle], | |
outputs=txt2vid_prompt | |
) | |
txt2vid_generate.click( | |
fn=generate_video_from_text, | |
inputs=[ | |
txt2vid_prompt, | |
txt2vid_enhance_toggle, | |
txt2vid_negative_prompt, | |
txt2vid_frame_rate, | |
*txt2vid_advanced[:3], | |
txt2vid_current_height, | |
txt2vid_current_width, | |
txt2vid_current_num_frames, | |
], | |
outputs=txt2vid_output, | |
concurrency_limit=1, | |
concurrency_id="generate_video", | |
queue=True, | |
) | |
img2vid_preset.change( | |
fn=preset_changed, | |
inputs=[img2vid_preset], | |
outputs=[ | |
img2vid_current_height, | |
img2vid_current_width, | |
img2vid_current_num_frames, | |
*img2vid_advanced[3:] | |
] | |
) | |
img2vid_enhance_toggle.change( | |
fn=update_prompt_i2v, | |
inputs=[img2vid_prompt, img2vid_enhance_toggle], | |
outputs=img2vid_prompt | |
) | |
img2vid_generate.click( | |
fn=generate_video_from_image, | |
inputs=[ | |
img2vid_image, | |
img2vid_prompt, | |
img2vid_enhance_toggle, | |
img2vid_negative_prompt, | |
img2vid_frame_rate, | |
*img2vid_advanced[:3], | |
img2vid_current_height, | |
img2vid_current_width, | |
img2vid_current_num_frames, | |
], | |
outputs=img2vid_output, | |
concurrency_limit=1, | |
concurrency_id="generate_video", | |
queue=True, | |
) | |
# Scenario tab event handlers | |
analyze_btn.click( | |
fn=analyze_scenario, | |
inputs=[scenario_input], | |
outputs=[ | |
section1_prompt, section2_prompt, section3_prompt, | |
section4_prompt, section5_prompt | |
] | |
) | |
# ๊ฐ ์น์ ์ ํ๋กฌํํธ ์ฌ์์ฑ ์ด๋ฒคํธ ํธ๋ค๋ฌ ์ถ๊ฐ | |
section1_regenerate.click( | |
fn=lambda x: generate_single_section_prompt(x, 1), | |
inputs=[scenario_input], | |
outputs=section1_prompt | |
) | |
section2_regenerate.click( | |
fn=lambda x: generate_single_section_prompt(x, 2), | |
inputs=[scenario_input], | |
outputs=section2_prompt | |
) | |
section3_regenerate.click( | |
fn=lambda x: generate_single_section_prompt(x, 3), | |
inputs=[scenario_input], | |
outputs=section3_prompt | |
) | |
section4_regenerate.click( | |
fn=lambda x: generate_single_section_prompt(x, 4), | |
inputs=[scenario_input], | |
outputs=section4_prompt | |
) | |
section5_regenerate.click( | |
fn=lambda x: generate_single_section_prompt(x, 5), | |
inputs=[scenario_input], | |
outputs=section5_prompt | |
) | |
# ์น์ ์์ฑ ์ด๋ฒคํธ ํธ๋ค๋ฌ | |
section1_generate.click( | |
fn=generate_section_video, | |
inputs=[section1_prompt, scenario_preset], | |
outputs=section1_video, | |
api_name=f"generate_section1" | |
) | |
section2_generate.click( | |
fn=lambda p, pr: generate_section_video(p, pr, 2), | |
inputs=[section2_prompt, scenario_preset], | |
outputs=section2_video, | |
api_name=f"generate_section2" | |
) | |
section3_generate.click( | |
fn=lambda p, pr: generate_section_video(p, pr, 3), | |
inputs=[section3_prompt, scenario_preset], | |
outputs=section3_video, | |
api_name=f"generate_section3" | |
) | |
section4_generate.click( | |
fn=lambda p, pr: generate_section_video(p, pr, 4), | |
inputs=[section4_prompt, scenario_preset], | |
outputs=section4_video, | |
api_name=f"generate_section4" | |
) | |
section5_generate.click( | |
fn=lambda p, pr: generate_section_video(p, pr, 5), | |
inputs=[section5_prompt, scenario_preset], | |
outputs=section5_video, | |
api_name=f"generate_section5" | |
) | |
# ์ด๋ฒคํธ ํธ๋ค๋ฌ ์ถ๊ฐ | |
merge_videos_btn.click( | |
fn=merge_section_videos, | |
inputs=[ | |
section1_video, | |
section2_video, | |
section3_video, | |
section4_video, | |
section5_video | |
], | |
outputs=merged_video_output | |
) | |
if __name__ == "__main__": | |
iface.queue(max_size=64, default_concurrency_limit=1, api_open=False).launch( | |
share=True, show_api=False | |
) |