init

README.md

@@ -10,4 +10,15 @@ pinned: false
 license: mit
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+
+
+## setup locally
+conda create -n matgen python=3.11
+conda activate matgen
+pip install diffusers["torch"] transformers accelerate xformers
+pip install gradio
+pip install controlnet-aux
+
+## local authen
+huggingface-cli login

app.py

@@ -1,156 +1,72 @@
-import gradio as gr
-import numpy as np
-import random
+#!/usr/bin/env python
 
-# import spaces #[uncomment to use ZeroGPU]
-from diffusers import DiffusionPipeline
+import gradio as gr
 import torch
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
-model_repo_id = "stabilityai/sdxl-turbo"  # Replace to the model you would like to use
-
-if torch.cuda.is_available():
-    torch_dtype = torch.float16
-else:
-    torch_dtype = torch.float32
-
-pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
-pipe = pipe.to(device)
-
-MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
-
-
-# @spaces.GPU #[uncomment to use ZeroGPU]
-def infer(
-    prompt,
-    negative_prompt,
-    seed,
-    randomize_seed,
-    width,
-    height,
-    guidance_scale,
-    num_inference_steps,
-    progress=gr.Progress(track_tqdm=True),
-):
-    if randomize_seed:
-        seed = random.randint(0, MAX_SEED)
+from app_canny import create_demo as create_demo_canny
 
-    generator = torch.Generator().manual_seed(seed)
+from model import Model
+from settings import ALLOW_CHANGING_BASE_MODEL, DEFAULT_MODEL_ID, SHOW_DUPLICATE_BUTTON
 
-    image = pipe(
-        prompt=prompt,
-        negative_prompt=negative_prompt,
-        guidance_scale=guidance_scale,
-        num_inference_steps=num_inference_steps,
-        width=width,
-        height=height,
-        generator=generator,
-    ).images[0]
+DESCRIPTION = "# Material Authoring Demo v0.1"
 
-    return image, seed
+if not torch.cuda.is_available():
+    DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
 
+model = Model(base_model_id=DEFAULT_MODEL_ID, task_name="Canny")
 
-examples = [
-    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
-    "An astronaut riding a green horse",
-    "A delicious ceviche cheesecake slice",
-]
-
-css = """
-#col-container {
-    margin: 0 auto;
-    max-width: 640px;
-}
-"""
+with gr.Blocks() as demo:
+    gr.Markdown(DESCRIPTION)
+    gr.DuplicateButton(
+        value="Duplicate Space for private use",
+        elem_id="duplicate-button",
+        visible=SHOW_DUPLICATE_BUTTON,
+    )
 
-with gr.Blocks(css=css) as demo:
-    with gr.Column(elem_id="col-container"):
-        gr.Markdown(" # Text-to-Image Gradio Template")
+    with gr.Tabs():
+        with gr.Tab("Canny"):
+            create_demo_canny(model.process_canny)
+        with gr.Tab("Texnet"):
+            create_demo_canny(model.process_canny)
+        with gr.Tab("Matnet"):
+            create_demo_canny(model.process_canny)
 
+    with gr.Accordion(label="Base model", open=False):
         with gr.Row():
-            prompt = gr.Text(
-                label="Prompt",
-                show_label=False,
-                max_lines=1,
-                placeholder="Enter your prompt",
-                container=False,
-            )
-
-            run_button = gr.Button("Run", scale=0, variant="primary")
-
-        result = gr.Image(label="Result", show_label=False)
-
-        with gr.Accordion("Advanced Settings", open=False):
-            negative_prompt = gr.Text(
-                label="Negative prompt",
-                max_lines=1,
-                placeholder="Enter a negative prompt",
-                visible=False,
-            )
-
-            seed = gr.Slider(
-                label="Seed",
-                minimum=0,
-                maximum=MAX_SEED,
-                step=1,
-                value=0,
-            )
-
-            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
-
-            with gr.Row():
-                width = gr.Slider(
-                    label="Width",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
-
-                height = gr.Slider(
-                    label="Height",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024,  # Replace with defaults that work for your model
-                )
-
-            with gr.Row():
-                guidance_scale = gr.Slider(
-                    label="Guidance scale",
-                    minimum=0.0,
-                    maximum=10.0,
-                    step=0.1,
-                    value=0.0,  # Replace with defaults that work for your model
-                )
-
-                num_inference_steps = gr.Slider(
-                    label="Number of inference steps",
-                    minimum=1,
-                    maximum=50,
-                    step=1,
-                    value=2,  # Replace with defaults that work for your model
+            with gr.Column(scale=5):
+                current_base_model = gr.Text(label="Current base model")
+            with gr.Column(scale=1):
+                check_base_model_button = gr.Button("Check current base model")
+        with gr.Row():
+            with gr.Column(scale=5):
+                new_base_model_id = gr.Text(
+                    label="New base model",
+                    max_lines=1,
+                    placeholder="stable-diffusion-v1-5/stable-diffusion-v1-5",
+                    info="The base model must be compatible with Stable Diffusion v1.5.",
+                    interactive=ALLOW_CHANGING_BASE_MODEL,
                 )
+            with gr.Column(scale=1):
+                change_base_model_button = gr.Button("Change base model", interactive=ALLOW_CHANGING_BASE_MODEL)
+        if not ALLOW_CHANGING_BASE_MODEL:
+            gr.Markdown(
+                """The base model is not allowed to be changed in this Space so as not to slow down the demo, but it can be changed if you duplicate the Space."""
+            )
 
-        gr.Examples(examples=examples, inputs=[prompt])
+    check_base_model_button.click(
+        fn=lambda: model.base_model_id,
+        outputs=current_base_model,
+        queue=False,
+        api_name="check_base_model",
+    )
     gr.on(
-        triggers=[run_button.click, prompt.submit],
-        fn=infer,
-        inputs=[
-            prompt,
-            negative_prompt,
-            seed,
-            randomize_seed,
-            width,
-            height,
-            guidance_scale,
-            num_inference_steps,
-        ],
-        outputs=[result, seed],
+        triggers=[new_base_model_id.submit, change_base_model_button.click],
+        fn=model.set_base_model,
+        inputs=new_base_model_id,
+        outputs=current_base_model,
+        api_name=False,
+        concurrency_id="main",
     )
 
 if __name__ == "__main__":
-    demo.launch()
-
-    print()
+    demo.queue(max_size=20).launch()

app_canny.py

@@ -0,0 +1,83 @@
+#!/usr/bin/env python
+
+import gradio as gr
+
+from settings import (
+    DEFAULT_IMAGE_RESOLUTION,
+    DEFAULT_NUM_IMAGES,
+    MAX_IMAGE_RESOLUTION,
+    MAX_NUM_IMAGES,
+    MAX_SEED,
+)
+from utils import randomize_seed_fn
+
+
+def create_demo(process):
+    with gr.Blocks() as demo:
+        with gr.Row():
+            with gr.Column():
+                image = gr.Image()
+                prompt = gr.Textbox(label="Prompt", submit_btn=True)
+                with gr.Accordion("Advanced options", open=False):
+                    num_samples = gr.Slider(
+                        label="Number of images", minimum=1, maximum=MAX_NUM_IMAGES, value=DEFAULT_NUM_IMAGES, step=1
+                    )
+                    image_resolution = gr.Slider(
+                        label="Image resolution",
+                        minimum=256,
+                        maximum=MAX_IMAGE_RESOLUTION,
+                        value=DEFAULT_IMAGE_RESOLUTION,
+                        step=256,
+                    )
+                    canny_low_threshold = gr.Slider(
+                        label="Canny low threshold", minimum=1, maximum=255, value=100, step=1
+                    )
+                    canny_high_threshold = gr.Slider(
+                        label="Canny high threshold", minimum=1, maximum=255, value=200, step=1
+                    )
+                    num_steps = gr.Slider(label="Number of steps", minimum=1, maximum=100, value=20, step=1)
+                    guidance_scale = gr.Slider(label="Guidance scale", minimum=0.1, maximum=30.0, value=9.0, step=0.1)
+                    seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0)
+                    randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
+                    a_prompt = gr.Textbox(label="Additional prompt", value="best quality, extremely detailed")
+                    n_prompt = gr.Textbox(
+                        label="Negative prompt",
+                        value="longbody, lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality",
+                    )
+            with gr.Column():
+                result = gr.Gallery(label="Output", show_label=False, columns=2, object_fit="scale-down")
+        inputs = [
+            image,
+            prompt,
+            a_prompt,
+            n_prompt,
+            num_samples,
+            image_resolution,
+            num_steps,
+            guidance_scale,
+            seed,
+            canny_low_threshold,
+            canny_high_threshold,
+        ]
+        prompt.submit(
+            fn=randomize_seed_fn,
+            inputs=[seed, randomize_seed],
+            outputs=seed,
+            queue=False,
+            api_name=False,
+        ).then(
+            fn=process,
+            inputs=inputs,
+            outputs=result,
+            api_name="canny",
+            concurrency_id="main",
+        )
+    return demo
+
+
+if __name__ == "__main__":
+    from model import Model
+
+    model = Model(task_name="Canny")
+    demo = create_demo(model.process_canny)
+    demo.queue().launch()

app_matnet.py

@@ -0,0 +1,83 @@
+#!/usr/bin/env python
+
+import gradio as gr
+
+from settings import (
+    DEFAULT_IMAGE_RESOLUTION,
+    DEFAULT_NUM_IMAGES,
+    MAX_IMAGE_RESOLUTION,
+    MAX_NUM_IMAGES,
+    MAX_SEED,
+)
+from utils import randomize_seed_fn
+
+
+def create_demo(process):
+    with gr.Blocks() as demo:
+        with gr.Row():
+            with gr.Column():
+                image = gr.Image()
+                prompt = gr.Textbox(label="Prompt", submit_btn=True)
+                with gr.Accordion("Advanced options", open=False):
+                    num_samples = gr.Slider(
+                        label="Number of images", minimum=1, maximum=MAX_NUM_IMAGES, value=DEFAULT_NUM_IMAGES, step=1
+                    )
+                    image_resolution = gr.Slider(
+                        label="Image resolution",
+                        minimum=256,
+                        maximum=MAX_IMAGE_RESOLUTION,
+                        value=DEFAULT_IMAGE_RESOLUTION,
+                        step=256,
+                    )
+                    canny_low_threshold = gr.Slider(
+                        label="Canny low threshold", minimum=1, maximum=255, value=100, step=1
+                    )
+                    canny_high_threshold = gr.Slider(
+                        label="Canny high threshold", minimum=1, maximum=255, value=200, step=1
+                    )
+                    num_steps = gr.Slider(label="Number of steps", minimum=1, maximum=100, value=20, step=1)
+                    guidance_scale = gr.Slider(label="Guidance scale", minimum=0.1, maximum=30.0, value=9.0, step=0.1)
+                    seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0)
+                    randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
+                    a_prompt = gr.Textbox(label="Additional prompt", value="best quality, extremely detailed")
+                    n_prompt = gr.Textbox(
+                        label="Negative prompt",
+                        value="longbody, lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality",
+                    )
+            with gr.Column():
+                result = gr.Gallery(label="Output", show_label=False, columns=2, object_fit="scale-down")
+        inputs = [
+            image,
+            prompt,
+            a_prompt,
+            n_prompt,
+            num_samples,
+            image_resolution,
+            num_steps,
+            guidance_scale,
+            seed,
+            canny_low_threshold,
+            canny_high_threshold,
+        ]
+        prompt.submit(
+            fn=randomize_seed_fn,
+            inputs=[seed, randomize_seed],
+            outputs=seed,
+            queue=False,
+            api_name=False,
+        ).then(
+            fn=process,
+            inputs=inputs,
+            outputs=result,
+            api_name="canny",
+            concurrency_id="main",
+        )
+    return demo
+
+
+if __name__ == "__main__":
+    from model import Model
+
+    model = Model(task_name="Canny")
+    demo = create_demo(model.process_canny)
+    demo.queue().launch()

app_texnet.py

@@ -0,0 +1,83 @@
+#!/usr/bin/env python
+
+import gradio as gr
+
+from settings import (
+    DEFAULT_IMAGE_RESOLUTION,
+    DEFAULT_NUM_IMAGES,
+    MAX_IMAGE_RESOLUTION,
+    MAX_NUM_IMAGES,
+    MAX_SEED,
+)
+from utils import randomize_seed_fn
+
+
+def create_demo(process):
+    with gr.Blocks() as demo:
+        with gr.Row():
+            with gr.Column():
+                image = gr.Image()
+                prompt = gr.Textbox(label="Prompt", submit_btn=True)
+                with gr.Accordion("Advanced options", open=False):
+                    num_samples = gr.Slider(
+                        label="Number of images", minimum=1, maximum=MAX_NUM_IMAGES, value=DEFAULT_NUM_IMAGES, step=1
+                    )
+                    image_resolution = gr.Slider(
+                        label="Image resolution",
+                        minimum=256,
+                        maximum=MAX_IMAGE_RESOLUTION,
+                        value=DEFAULT_IMAGE_RESOLUTION,
+                        step=256,
+                    )
+                    canny_low_threshold = gr.Slider(
+                        label="Canny low threshold", minimum=1, maximum=255, value=100, step=1
+                    )
+                    canny_high_threshold = gr.Slider(
+                        label="Canny high threshold", minimum=1, maximum=255, value=200, step=1
+                    )
+                    num_steps = gr.Slider(label="Number of steps", minimum=1, maximum=100, value=20, step=1)
+                    guidance_scale = gr.Slider(label="Guidance scale", minimum=0.1, maximum=30.0, value=9.0, step=0.1)
+                    seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0)
+                    randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
+                    a_prompt = gr.Textbox(label="Additional prompt", value="best quality, extremely detailed")
+                    n_prompt = gr.Textbox(
+                        label="Negative prompt",
+                        value="longbody, lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality",
+                    )
+            with gr.Column():
+                result = gr.Gallery(label="Output", show_label=False, columns=2, object_fit="scale-down")
+        inputs = [
+            image,
+            prompt,
+            a_prompt,
+            n_prompt,
+            num_samples,
+            image_resolution,
+            num_steps,
+            guidance_scale,
+            seed,
+            canny_low_threshold,
+            canny_high_threshold,
+        ]
+        prompt.submit(
+            fn=randomize_seed_fn,
+            inputs=[seed, randomize_seed],
+            outputs=seed,
+            queue=False,
+            api_name=False,
+        ).then(
+            fn=process,
+            inputs=inputs,
+            outputs=result,
+            api_name="canny",
+            concurrency_id="main",
+        )
+    return demo
+
+
+if __name__ == "__main__":
+    from model import Model
+
+    model = Model(task_name="Canny")
+    demo = create_demo(model.process_canny)
+    demo.queue().launch()

cv_utils.py

@@ -0,0 +1,17 @@
+import cv2
+import numpy as np
+
+
+def resize_image(input_image, resolution, interpolation=None):
+    H, W, C = input_image.shape
+    H = float(H)
+    W = float(W)
+    k = float(resolution) / max(H, W)
+    H *= k
+    W *= k
+    H = int(np.round(H / 64.0)) * 64
+    W = int(np.round(W / 64.0)) * 64
+    if interpolation is None:
+        interpolation = cv2.INTER_LANCZOS4 if k > 1 else cv2.INTER_AREA
+    img = cv2.resize(input_image, (W, H), interpolation=interpolation)
+    return img

depth_estimator.py

@@ -0,0 +1,25 @@
+import numpy as np
+import PIL.Image
+from controlnet_aux.util import HWC3
+from transformers import pipeline
+
+from cv_utils import resize_image
+
+
+class DepthEstimator:
+    def __init__(self):
+        self.model = pipeline("depth-estimation")
+
+    def __call__(self, image: np.ndarray, **kwargs) -> PIL.Image.Image:
+        detect_resolution = kwargs.pop("detect_resolution", 512)
+        image_resolution = kwargs.pop("image_resolution", 512)
+        image = np.array(image)
+        image = HWC3(image)
+        image = resize_image(image, resolution=detect_resolution)
+        image = PIL.Image.fromarray(image)
+        image = self.model(image)
+        image = image["depth"]
+        image = np.array(image)
+        image = HWC3(image)
+        image = resize_image(image, resolution=image_resolution)
+        return PIL.Image.fromarray(image)

image_segmentor.py

@@ -0,0 +1,33 @@
+import cv2
+import numpy as np
+import PIL.Image
+import torch
+from controlnet_aux.util import HWC3, ade_palette
+from transformers import AutoImageProcessor, UperNetForSemanticSegmentation
+
+from cv_utils import resize_image
+
+
+class ImageSegmentor:
+    def __init__(self):
+        self.image_processor = AutoImageProcessor.from_pretrained("openmmlab/upernet-convnext-small")
+        self.image_segmentor = UperNetForSemanticSegmentation.from_pretrained("openmmlab/upernet-convnext-small")
+
+    @torch.inference_mode()
+    def __call__(self, image: np.ndarray, **kwargs) -> PIL.Image.Image:
+        detect_resolution = kwargs.pop("detect_resolution", 512)
+        image_resolution = kwargs.pop("image_resolution", 512)
+        image = HWC3(image)
+        image = resize_image(image, resolution=detect_resolution)
+        image = PIL.Image.fromarray(image)
+
+        pixel_values = self.image_processor(image, return_tensors="pt").pixel_values
+        outputs = self.image_segmentor(pixel_values)
+        seg = self.image_processor.post_process_semantic_segmentation(outputs, target_sizes=[image.size[::-1]])[0]
+        color_seg = np.zeros((seg.shape[0], seg.shape[1], 3), dtype=np.uint8)
+        for label, color in enumerate(ade_palette()):
+            color_seg[seg == label, :] = color
+        color_seg = color_seg.astype(np.uint8)
+
+        color_seg = resize_image(color_seg, resolution=image_resolution, interpolation=cv2.INTER_NEAREST)
+        return PIL.Image.fromarray(color_seg)

model.py

@@ -0,0 +1,670 @@
+import gc
+
+import numpy as np
+import PIL.Image
+import torch
+from controlnet_aux.util import HWC3
+from diffusers import (
+    ControlNetModel,
+    DiffusionPipeline,
+    StableDiffusionControlNetPipeline,
+    UniPCMultistepScheduler,
+)
+
+from cv_utils import resize_image
+from preprocessor import Preprocessor
+from settings import MAX_IMAGE_RESOLUTION, MAX_NUM_IMAGES
+
+CONTROLNET_MODEL_IDS = {
+    "Openpose": "lllyasviel/control_v11p_sd15_openpose",
+    "Canny": "lllyasviel/control_v11p_sd15_canny",
+    "MLSD": "lllyasviel/control_v11p_sd15_mlsd",
+    "scribble": "lllyasviel/control_v11p_sd15_scribble",
+    "softedge": "lllyasviel/control_v11p_sd15_softedge",
+    "segmentation": "lllyasviel/control_v11p_sd15_seg",
+    "depth": "lllyasviel/control_v11f1p_sd15_depth",
+    "NormalBae": "lllyasviel/control_v11p_sd15_normalbae",
+    "lineart": "lllyasviel/control_v11p_sd15_lineart",
+    "lineart_anime": "lllyasviel/control_v11p_sd15s2_lineart_anime",
+    "shuffle": "lllyasviel/control_v11e_sd15_shuffle",
+    "ip2p": "lllyasviel/control_v11e_sd15_ip2p",
+    "inpaint": "lllyasviel/control_v11e_sd15_inpaint",
+}
+
+
+def download_all_controlnet_weights() -> None:
+    for model_id in CONTROLNET_MODEL_IDS.values():
+        ControlNetModel.from_pretrained(model_id)
+
+
+class Model:
+    def __init__(
+        self, base_model_id: str = "stable-diffusion-v1-5/stable-diffusion-v1-5", task_name: str = "Canny"
+    ) -> None:
+        self.device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+        self.base_model_id = ""
+        self.task_name = ""
+        self.pipe = self.load_pipe(base_model_id, task_name)
+        self.preprocessor = Preprocessor()
+
+    def load_pipe(self, base_model_id: str, task_name: str) -> DiffusionPipeline:
+        if (
+            base_model_id == self.base_model_id
+            and task_name == self.task_name
+            and hasattr(self, "pipe")
+            and self.pipe is not None
+        ):
+            return self.pipe
+        model_id = CONTROLNET_MODEL_IDS[task_name]
+        controlnet = ControlNetModel.from_pretrained(model_id, torch_dtype=torch.float16)
+        pipe = StableDiffusionControlNetPipeline.from_pretrained(
+            base_model_id, safety_checker=None, controlnet=controlnet, torch_dtype=torch.float16
+        )
+        pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
+        if self.device.type == "cuda":
+            pipe.enable_xformers_memory_efficient_attention()
+        pipe.to(self.device)
+        torch.cuda.empty_cache()
+        gc.collect()
+        self.base_model_id = base_model_id
+        self.task_name = task_name
+        return pipe
+
+    def set_base_model(self, base_model_id: str) -> str:
+        if not base_model_id or base_model_id == self.base_model_id:
+            return self.base_model_id
+        del self.pipe
+        torch.cuda.empty_cache()
+        gc.collect()
+        try:
+            self.pipe = self.load_pipe(base_model_id, self.task_name)
+        except Exception:  # noqa: BLE001
+            self.pipe = self.load_pipe(self.base_model_id, self.task_name)
+        return self.base_model_id
+
+    def load_controlnet_weight(self, task_name: str) -> None:
+        if task_name == self.task_name:
+            return
+        if self.pipe is not None and hasattr(self.pipe, "controlnet"):
+            del self.pipe.controlnet
+        torch.cuda.empty_cache()
+        gc.collect()
+        model_id = CONTROLNET_MODEL_IDS[task_name]
+        controlnet = ControlNetModel.from_pretrained(model_id, torch_dtype=torch.float16)
+        controlnet.to(self.device)
+        torch.cuda.empty_cache()
+        gc.collect()
+        self.pipe.controlnet = controlnet
+        self.task_name = task_name
+
+    def get_prompt(self, prompt: str, additional_prompt: str) -> str:
+        return additional_prompt if not prompt else f"{prompt}, {additional_prompt}"
+
+    @torch.autocast("cuda")
+    def run_pipe(
+        self,
+        prompt: str,
+        negative_prompt: str,
+        control_image: PIL.Image.Image,
+        num_images: int,
+        num_steps: int,
+        guidance_scale: float,
+        seed: int,
+    ) -> list[PIL.Image.Image]:
+        generator = torch.Generator().manual_seed(seed)
+        return self.pipe(
+            prompt=prompt,
+            negative_prompt=negative_prompt,
+            guidance_scale=guidance_scale,
+            num_images_per_prompt=num_images,
+            num_inference_steps=num_steps,
+            generator=generator,
+            image=control_image,
+        ).images
+
+    @torch.inference_mode()
+    def process_canny(
+        self,
+        image: np.ndarray,
+        prompt: str,
+        additional_prompt: str,
+        negative_prompt: str,
+        num_images: int,
+        image_resolution: int,
+        num_steps: int,
+        guidance_scale: float,
+        seed: int,
+        low_threshold: int,
+        high_threshold: int,
+    ) -> list[PIL.Image.Image]:
+        if image is None:
+            raise ValueError
+        if image_resolution > MAX_IMAGE_RESOLUTION:
+            raise ValueError
+        if num_images > MAX_NUM_IMAGES:
+            raise ValueError
+
+        self.preprocessor.load("Canny")
+        control_image = self.preprocessor(
+            image=image, low_threshold=low_threshold, high_threshold=high_threshold, detect_resolution=image_resolution
+        )
+
+        self.load_controlnet_weight("Canny")
+        results = self.run_pipe(
+            prompt=self.get_prompt(prompt, additional_prompt),
+            negative_prompt=negative_prompt,
+            control_image=control_image,
+            num_images=num_images,
+            num_steps=num_steps,
+            guidance_scale=guidance_scale,
+            seed=seed,
+        )
+        return [control_image, *results]
+
+    @torch.inference_mode()
+    def process_mlsd(
+        self,
+        image: np.ndarray,
+        prompt: str,
+        additional_prompt: str,
+        negative_prompt: str,
+        num_images: int,
+        image_resolution: int,
+        preprocess_resolution: int,
+        num_steps: int,
+        guidance_scale: float,
+        seed: int,
+        value_threshold: float,
+        distance_threshold: float,
+    ) -> list[PIL.Image.Image]:
+        if image is None:
+            raise ValueError
+        if image_resolution > MAX_IMAGE_RESOLUTION:
+            raise ValueError
+        if num_images > MAX_NUM_IMAGES:
+            raise ValueError
+
+        self.preprocessor.load("MLSD")
+        control_image = self.preprocessor(
+            image=image,
+            image_resolution=image_resolution,
+            detect_resolution=preprocess_resolution,
+            thr_v=value_threshold,
+            thr_d=distance_threshold,
+        )
+        self.load_controlnet_weight("MLSD")
+        results = self.run_pipe(
+            prompt=self.get_prompt(prompt, additional_prompt),
+            negative_prompt=negative_prompt,
+            control_image=control_image,
+            num_images=num_images,
+            num_steps=num_steps,
+            guidance_scale=guidance_scale,
+            seed=seed,
+        )
+        return [control_image, *results]
+
+    @torch.inference_mode()
+    def process_scribble(
+        self,
+        image: np.ndarray,
+        prompt: str,
+        additional_prompt: str,
+        negative_prompt: str,
+        num_images: int,
+        image_resolution: int,
+        preprocess_resolution: int,
+        num_steps: int,
+        guidance_scale: float,
+        seed: int,
+        preprocessor_name: str,
+    ) -> list[PIL.Image.Image]:
+        if image is None:
+            raise ValueError
+        if image_resolution > MAX_IMAGE_RESOLUTION:
+            raise ValueError
+        if num_images > MAX_NUM_IMAGES:
+            raise ValueError
+
+        if preprocessor_name == "None":
+            image = HWC3(image)
+            image = resize_image(image, resolution=image_resolution)
+            control_image = PIL.Image.fromarray(image)
+        elif preprocessor_name == "HED":
+            self.preprocessor.load(preprocessor_name)
+            control_image = self.preprocessor(
+                image=image,
+                image_resolution=image_resolution,
+                detect_resolution=preprocess_resolution,
+                scribble=False,
+            )
+        elif preprocessor_name == "PidiNet":
+            self.preprocessor.load(preprocessor_name)
+            control_image = self.preprocessor(
+                image=image,
+                image_resolution=image_resolution,
+                detect_resolution=preprocess_resolution,
+                safe=False,
+            )
+        self.load_controlnet_weight("scribble")
+        results = self.run_pipe(
+            prompt=self.get_prompt(prompt, additional_prompt),
+            negative_prompt=negative_prompt,
+            control_image=control_image,
+            num_images=num_images,
+            num_steps=num_steps,
+            guidance_scale=guidance_scale,
+            seed=seed,
+        )
+        return [control_image, *results]
+
+    @torch.inference_mode()
+    def process_scribble_interactive(
+        self,
+        image_and_mask: dict[str, np.ndarray | list[np.ndarray]] | None,
+        prompt: str,
+        additional_prompt: str,
+        negative_prompt: str,
+        num_images: int,
+        image_resolution: int,
+        num_steps: int,
+        guidance_scale: float,
+        seed: int,
+    ) -> list[PIL.Image.Image]:
+        if image_and_mask is None:
+            raise ValueError
+        if image_resolution > MAX_IMAGE_RESOLUTION:
+            raise ValueError
+        if num_images > MAX_NUM_IMAGES:
+            raise ValueError
+
+        image = 255 - image_and_mask["composite"]  # type: ignore
+        image = HWC3(image)
+        image = resize_image(image, resolution=image_resolution)
+        control_image = PIL.Image.fromarray(image)
+
+        self.load_controlnet_weight("scribble")
+        results = self.run_pipe(
+            prompt=self.get_prompt(prompt, additional_prompt),
+            negative_prompt=negative_prompt,
+            control_image=control_image,
+            num_images=num_images,
+            num_steps=num_steps,
+            guidance_scale=guidance_scale,
+            seed=seed,
+        )
+        return [control_image, *results]
+
+    @torch.inference_mode()
+    def process_softedge(
+        self,
+        image: np.ndarray,
+        prompt: str,
+        additional_prompt: str,
+        negative_prompt: str,
+        num_images: int,
+        image_resolution: int,
+        preprocess_resolution: int,
+        num_steps: int,
+        guidance_scale: float,
+        seed: int,
+        preprocessor_name: str,
+    ) -> list[PIL.Image.Image]:
+        if image is None:
+            raise ValueError
+        if image_resolution > MAX_IMAGE_RESOLUTION:
+            raise ValueError
+        if num_images > MAX_NUM_IMAGES:
+            raise ValueError
+
+        if preprocessor_name == "None":
+            image = HWC3(image)
+            image = resize_image(image, resolution=image_resolution)
+            control_image = PIL.Image.fromarray(image)
+        elif preprocessor_name in ["HED", "HED safe"]:
+            safe = "safe" in preprocessor_name
+            self.preprocessor.load("HED")
+            control_image = self.preprocessor(
+                image=image,
+                image_resolution=image_resolution,
+                detect_resolution=preprocess_resolution,
+                scribble=safe,
+            )
+        elif preprocessor_name in ["PidiNet", "PidiNet safe"]:
+            safe = "safe" in preprocessor_name
+            self.preprocessor.load("PidiNet")
+            control_image = self.preprocessor(
+                image=image,
+                image_resolution=image_resolution,
+                detect_resolution=preprocess_resolution,
+                safe=safe,
+            )
+        else:
+            raise ValueError
+        self.load_controlnet_weight("softedge")
+        results = self.run_pipe(
+            prompt=self.get_prompt(prompt, additional_prompt),
+            negative_prompt=negative_prompt,
+            control_image=control_image,
+            num_images=num_images,
+            num_steps=num_steps,
+            guidance_scale=guidance_scale,
+            seed=seed,
+        )
+        return [control_image, *results]
+
+    @torch.inference_mode()
+    def process_openpose(
+        self,
+        image: np.ndarray,
+        prompt: str,
+        additional_prompt: str,
+        negative_prompt: str,
+        num_images: int,
+        image_resolution: int,
+        preprocess_resolution: int,
+        num_steps: int,
+        guidance_scale: float,
+        seed: int,
+        preprocessor_name: str,
+    ) -> list[PIL.Image.Image]:
+        if image is None:
+            raise ValueError
+        if image_resolution > MAX_IMAGE_RESOLUTION:
+            raise ValueError
+        if num_images > MAX_NUM_IMAGES:
+            raise ValueError
+
+        if preprocessor_name == "None":
+            image = HWC3(image)
+            image = resize_image(image, resolution=image_resolution)
+            control_image = PIL.Image.fromarray(image)
+        else:
+            self.preprocessor.load("Openpose")
+            control_image = self.preprocessor(
+                image=image,
+                image_resolution=image_resolution,
+                detect_resolution=preprocess_resolution,
+                hand_and_face=True,
+            )
+        self.load_controlnet_weight("Openpose")
+        results = self.run_pipe(
+            prompt=self.get_prompt(prompt, additional_prompt),
+            negative_prompt=negative_prompt,
+            control_image=control_image,
+            num_images=num_images,
+            num_steps=num_steps,
+            guidance_scale=guidance_scale,
+            seed=seed,
+        )
+        return [control_image, *results]
+
+    @torch.inference_mode()
+    def process_segmentation(
+        self,
+        image: np.ndarray,
+        prompt: str,
+        additional_prompt: str,
+        negative_prompt: str,
+        num_images: int,
+        image_resolution: int,
+        preprocess_resolution: int,
+        num_steps: int,
+        guidance_scale: float,
+        seed: int,
+        preprocessor_name: str,
+    ) -> list[PIL.Image.Image]:
+        if image is None:
+            raise ValueError
+        if image_resolution > MAX_IMAGE_RESOLUTION:
+            raise ValueError
+        if num_images > MAX_NUM_IMAGES:
+            raise ValueError
+
+        if preprocessor_name == "None":
+            image = HWC3(image)
+            image = resize_image(image, resolution=image_resolution)
+            control_image = PIL.Image.fromarray(image)
+        else:
+            self.preprocessor.load(preprocessor_name)
+            control_image = self.preprocessor(
+                image=image,
+                image_resolution=image_resolution,
+                detect_resolution=preprocess_resolution,
+            )
+        self.load_controlnet_weight("segmentation")
+        results = self.run_pipe(
+            prompt=self.get_prompt(prompt, additional_prompt),
+            negative_prompt=negative_prompt,
+            control_image=control_image,
+            num_images=num_images,
+            num_steps=num_steps,
+            guidance_scale=guidance_scale,
+            seed=seed,
+        )
+        return [control_image, *results]
+
+    @torch.inference_mode()
+    def process_depth(
+        self,
+        image: np.ndarray,
+        prompt: str,
+        additional_prompt: str,
+        negative_prompt: str,
+        num_images: int,
+        image_resolution: int,
+        preprocess_resolution: int,
+        num_steps: int,
+        guidance_scale: float,
+        seed: int,
+        preprocessor_name: str,
+    ) -> list[PIL.Image.Image]:
+        if image is None:
+            raise ValueError
+        if image_resolution > MAX_IMAGE_RESOLUTION:
+            raise ValueError
+        if num_images > MAX_NUM_IMAGES:
+            raise ValueError
+
+        if preprocessor_name == "None":
+            image = HWC3(image)
+            image = resize_image(image, resolution=image_resolution)
+            control_image = PIL.Image.fromarray(image)
+        else:
+            self.preprocessor.load(preprocessor_name)
+            control_image = self.preprocessor(
+                image=image,
+                image_resolution=image_resolution,
+                detect_resolution=preprocess_resolution,
+            )
+        self.load_controlnet_weight("depth")
+        results = self.run_pipe(
+            prompt=self.get_prompt(prompt, additional_prompt),
+            negative_prompt=negative_prompt,
+            control_image=control_image,
+            num_images=num_images,
+            num_steps=num_steps,
+            guidance_scale=guidance_scale,
+            seed=seed,
+        )
+        return [control_image, *results]
+
+    @torch.inference_mode()
+    def process_normal(
+        self,
+        image: np.ndarray,
+        prompt: str,
+        additional_prompt: str,
+        negative_prompt: str,
+        num_images: int,
+        image_resolution: int,
+        preprocess_resolution: int,
+        num_steps: int,
+        guidance_scale: float,
+        seed: int,
+        preprocessor_name: str,
+    ) -> list[PIL.Image.Image]:
+        if image is None:
+            raise ValueError
+        if image_resolution > MAX_IMAGE_RESOLUTION:
+            raise ValueError
+        if num_images > MAX_NUM_IMAGES:
+            raise ValueError
+
+        if preprocessor_name == "None":
+            image = HWC3(image)
+            image = resize_image(image, resolution=image_resolution)
+            control_image = PIL.Image.fromarray(image)
+        else:
+            self.preprocessor.load("NormalBae")
+            control_image = self.preprocessor(
+                image=image,
+                image_resolution=image_resolution,
+                detect_resolution=preprocess_resolution,
+            )
+        self.load_controlnet_weight("NormalBae")
+        results = self.run_pipe(
+            prompt=self.get_prompt(prompt, additional_prompt),
+            negative_prompt=negative_prompt,
+            control_image=control_image,
+            num_images=num_images,
+            num_steps=num_steps,
+            guidance_scale=guidance_scale,
+            seed=seed,
+        )
+        return [control_image, *results]
+
+    @torch.inference_mode()
+    def process_lineart(
+        self,
+        image: np.ndarray,
+        prompt: str,
+        additional_prompt: str,
+        negative_prompt: str,
+        num_images: int,
+        image_resolution: int,
+        preprocess_resolution: int,
+        num_steps: int,
+        guidance_scale: float,
+        seed: int,
+        preprocessor_name: str,
+    ) -> list[PIL.Image.Image]:
+        if image is None:
+            raise ValueError
+        if image_resolution > MAX_IMAGE_RESOLUTION:
+            raise ValueError
+        if num_images > MAX_NUM_IMAGES:
+            raise ValueError
+
+        if preprocessor_name in ["None", "None (anime)"]:
+            image = HWC3(image)
+            image = resize_image(image, resolution=image_resolution)
+            control_image = PIL.Image.fromarray(image)
+        elif preprocessor_name in ["Lineart", "Lineart coarse"]:
+            coarse = "coarse" in preprocessor_name
+            self.preprocessor.load("Lineart")
+            control_image = self.preprocessor(
+                image=image,
+                image_resolution=image_resolution,
+                detect_resolution=preprocess_resolution,
+                coarse=coarse,
+            )
+        elif preprocessor_name == "Lineart (anime)":
+            self.preprocessor.load("LineartAnime")
+            control_image = self.preprocessor(
+                image=image,
+                image_resolution=image_resolution,
+                detect_resolution=preprocess_resolution,
+            )
+        if "anime" in preprocessor_name:
+            self.load_controlnet_weight("lineart_anime")
+        else:
+            self.load_controlnet_weight("lineart")
+        results = self.run_pipe(
+            prompt=self.get_prompt(prompt, additional_prompt),
+            negative_prompt=negative_prompt,
+            control_image=control_image,
+            num_images=num_images,
+            num_steps=num_steps,
+            guidance_scale=guidance_scale,
+            seed=seed,
+        )
+        return [control_image, *results]
+
+    @torch.inference_mode()
+    def process_shuffle(
+        self,
+        image: np.ndarray,
+        prompt: str,
+        additional_prompt: str,
+        negative_prompt: str,
+        num_images: int,
+        image_resolution: int,
+        num_steps: int,
+        guidance_scale: float,
+        seed: int,
+        preprocessor_name: str,
+    ) -> list[PIL.Image.Image]:
+        if image is None:
+            raise ValueError
+        if image_resolution > MAX_IMAGE_RESOLUTION:
+            raise ValueError
+        if num_images > MAX_NUM_IMAGES:
+            raise ValueError
+
+        if preprocessor_name == "None":
+            image = HWC3(image)
+            image = resize_image(image, resolution=image_resolution)
+            control_image = PIL.Image.fromarray(image)
+        else:
+            self.preprocessor.load(preprocessor_name)
+            control_image = self.preprocessor(
+                image=image,
+                image_resolution=image_resolution,
+            )
+        self.load_controlnet_weight("shuffle")
+        results = self.run_pipe(
+            prompt=self.get_prompt(prompt, additional_prompt),
+            negative_prompt=negative_prompt,
+            control_image=control_image,
+            num_images=num_images,
+            num_steps=num_steps,
+            guidance_scale=guidance_scale,
+            seed=seed,
+        )
+        return [control_image, *results]
+
+    @torch.inference_mode()
+    def process_ip2p(
+        self,
+        image: np.ndarray,
+        prompt: str,
+        additional_prompt: str,
+        negative_prompt: str,
+        num_images: int,
+        image_resolution: int,
+        num_steps: int,
+        guidance_scale: float,
+        seed: int,
+    ) -> list[PIL.Image.Image]:
+        if image is None:
+            raise ValueError
+        if image_resolution > MAX_IMAGE_RESOLUTION:
+            raise ValueError
+        if num_images > MAX_NUM_IMAGES:
+            raise ValueError
+
+        image = HWC3(image)
+        image = resize_image(image, resolution=image_resolution)
+        control_image = PIL.Image.fromarray(image)
+        self.load_controlnet_weight("ip2p")
+        results = self.run_pipe(
+            prompt=self.get_prompt(prompt, additional_prompt),
+            negative_prompt=negative_prompt,
+            control_image=control_image,
+            num_images=num_images,
+            num_steps=num_steps,
+            guidance_scale=guidance_scale,
+            seed=seed,
+        )
+        return [control_image, *results]

preprocessor.py

@@ -0,0 +1,88 @@
+import gc
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    from collections.abc import Callable
+
+import numpy as np
+import PIL.Image
+import torch
+from controlnet_aux import (
+    CannyDetector,
+    ContentShuffleDetector,
+    HEDdetector,
+    LineartAnimeDetector,
+    LineartDetector,
+    MidasDetector,
+    MLSDdetector,
+    NormalBaeDetector,
+    OpenposeDetector,
+    PidiNetDetector,
+)
+from controlnet_aux.util import HWC3
+
+from cv_utils import resize_image
+from depth_estimator import DepthEstimator
+from image_segmentor import ImageSegmentor
+
+
+class Preprocessor:
+    MODEL_ID = "lllyasviel/Annotators"
+
+    def __init__(self) -> None:
+        self.model: Callable = None  # type: ignore
+        self.name = ""
+
+    def load(self, name: str) -> None:  # noqa: C901, PLR0912
+        if name == self.name:
+            return
+        if name == "HED":
+            self.model = HEDdetector.from_pretrained(self.MODEL_ID)
+        elif name == "Midas":
+            self.model = MidasDetector.from_pretrained(self.MODEL_ID)
+        elif name == "MLSD":
+            self.model = MLSDdetector.from_pretrained(self.MODEL_ID)
+        elif name == "Openpose":
+            self.model = OpenposeDetector.from_pretrained(self.MODEL_ID)
+        elif name == "PidiNet":
+            self.model = PidiNetDetector.from_pretrained(self.MODEL_ID)
+        elif name == "NormalBae":
+            self.model = NormalBaeDetector.from_pretrained(self.MODEL_ID)
+        elif name == "Lineart":
+            self.model = LineartDetector.from_pretrained(self.MODEL_ID)
+        elif name == "LineartAnime":
+            self.model = LineartAnimeDetector.from_pretrained(self.MODEL_ID)
+        elif name == "Canny":
+            self.model = CannyDetector()
+        elif name == "ContentShuffle":
+            self.model = ContentShuffleDetector()
+        elif name == "DPT":
+            self.model = DepthEstimator()
+        elif name == "UPerNet":
+            self.model = ImageSegmentor()
+        else:
+            raise ValueError
+        torch.cuda.empty_cache()
+        gc.collect()
+        self.name = name
+
+    def __call__(self, image: PIL.Image.Image, **kwargs) -> PIL.Image.Image:  # noqa: ANN003
+        if self.name == "Canny":
+            if "detect_resolution" in kwargs:
+                detect_resolution = kwargs.pop("detect_resolution")
+                image = np.array(image)
+                image = HWC3(image)
+                image = resize_image(image, resolution=detect_resolution)
+            image = self.model(image, **kwargs)
+            return PIL.Image.fromarray(image)
+        if self.name == "Midas":
+            detect_resolution = kwargs.pop("detect_resolution", 512)
+            image_resolution = kwargs.pop("image_resolution", 512)
+            image = np.array(image)
+            image = HWC3(image)
+            image = resize_image(image, resolution=detect_resolution)
+            image = self.model(image, **kwargs)
+            image = HWC3(image)
+            image = resize_image(image, resolution=image_resolution)
+            return PIL.Image.fromarray(image)
+        return self.model(image, **kwargs)

settings.py

@@ -0,0 +1,19 @@
+import os
+
+import numpy as np
+
+DEFAULT_MODEL_ID = os.getenv("DEFAULT_MODEL_ID", "stable-diffusion-v1-5/stable-diffusion-v1-5")
+
+MAX_NUM_IMAGES = int(os.getenv("MAX_NUM_IMAGES", "3"))
+DEFAULT_NUM_IMAGES = min(MAX_NUM_IMAGES, int(os.getenv("DEFAULT_NUM_IMAGES", "3")))
+MAX_IMAGE_RESOLUTION = int(os.getenv("MAX_IMAGE_RESOLUTION", "768"))
+DEFAULT_IMAGE_RESOLUTION = min(MAX_IMAGE_RESOLUTION, int(os.getenv("DEFAULT_IMAGE_RESOLUTION", "768")))
+
+ALLOW_CHANGING_BASE_MODEL = os.getenv("SPACE_ID") != "hysts/ControlNet-v1-1"
+SHOW_DUPLICATE_BUTTON = os.getenv("SHOW_DUPLICATE_BUTTON") == "1"
+
+MAX_SEED = np.iinfo(np.int32).max
+
+# setup CUDA
+if os.getenv("CUDA_VISIBLE_DEVICES") is None:
+    os.environ["CUDA_VISIBLE_DEVICES"] = "7"

utils.py

@@ -0,0 +1,9 @@
+import random
+
+from settings import MAX_SEED
+
+
+def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
+    if randomize_seed:
+        seed = random.randint(0, MAX_SEED)  # noqa: S311
+    return seed