import time from typing import cast from comfydeploy import ComfyDeploy import os import gradio as gr from gradio.components.image_editor import EditorValue from PIL import Image import requests import dotenv from gradio_imageslider import ImageSlider from io import BytesIO import base64 import numpy as np from loguru import logger dotenv.load_dotenv() API_KEY = os.environ.get("API_KEY") CLEANER_DEPLOYMENT_ID = os.environ.get( "CLEANER_DEPLOYMENT_ID", "CLEANER_DEPLOYMENT_ID_NOT_SET" ) MASKER_DEPLOYMENT_ID = os.environ.get( "MASKER_DEPLOYMENT_ID", "MASKER_DEPLOYMENT_ID_NOT_SET" ) if not API_KEY: raise ValueError("Please set API_KEY in your environment variables") if ( not CLEANER_DEPLOYMENT_ID or CLEANER_DEPLOYMENT_ID == "CLEANER_DEPLOYMENT_ID_NOT_SET" ): raise ValueError("Please set CLEANER_DEPLOYMENT_ID in your environment variables") client = ComfyDeploy(bearer_auth=API_KEY) def get_base64_from_image(image: Image.Image) -> str: buffered: BytesIO = BytesIO() image.save(buffered, format="PNG") return base64.b64encode(buffered.getvalue()).decode("utf-8") def compute_mask( image: Image.Image | str | None, progress: gr.Progress = gr.Progress() ) -> Image.Image | None: progress(0, desc="Preparing inputs...") if image is None: return None image = resize_image(image) image_base64 = get_base64_from_image(image) # Prepare inputs inputs: dict = { "input_image": f"data:image/png;base64,{image_base64}", "dilation_1_iterations": 20, "dilation_2_iterations": 30, "mask_blur_amount": 215, } # Call ComfyDeploy API try: result = client.run.create( request={"deployment_id": MASKER_DEPLOYMENT_ID, "inputs": inputs} ) if result and result.object: run_id: str = result.object.run_id progress(0, desc="Starting processing...") # Wait for the result while True: run_result = client.run.get(run_id=run_id) if not run_result.object: continue progress_value = run_result.object.progress or 0 status = run_result.object.live_status or "Cold starting..." progress(progress_value, desc=f"Status: {status}") if run_result.object.status == "success": for output in run_result.object.outputs or []: if output.data and output.data.images: image_url: str = output.data.images[0].url # Download and return the mask image response: requests.Response = requests.get(image_url) mask_image: Image.Image = Image.open( BytesIO(response.content) ) return mask_image return None elif run_result.object.status == "failed": logger.debug("Processing failed") return None time.sleep(1) # Wait for 1 second before checking the status again except Exception as e: logger.debug(f"Error: {e}") return None def create_editor_value(image: Image.Image, mask: Image.Image) -> EditorValue: # Convert image to numpy array image_np = np.array(image) # Resize mask to match image dimensions mask_resized = mask.resize((image_np.shape[1], image_np.shape[0]), Image.NEAREST) mask_np = np.array(mask_resized) # Ensure mask is grayscale if len(mask_np.shape) == 3: mask_np = mask_np[:, :, -1] # Create the layers array layers = np.zeros((image_np.shape[0], image_np.shape[1], 4), dtype=np.uint8) layers[:, :, 3] = mask_np # Create the composite image composite = np.zeros((image_np.shape[0], image_np.shape[1], 4), dtype=np.uint8) composite[:, :, :3] = image_np composite[:, :, 3] = np.where(mask_np == 255, 0, 255) return { "background": image_np, "layers": [layers], "composite": composite, } def run_masking( image: np.ndarray | Image.Image | str | None, progress: gr.Progress = gr.Progress(), profile: gr.OAuthProfile | None = None, ) -> EditorValue | None: if image is None: return None if profile is None: gr.Info("Please log in to process the image.") return None # Convert np.ndarray to Image.Image if isinstance(image, np.ndarray): image = Image.fromarray(image) elif isinstance(image, str): image = Image.open(image) mask = compute_mask(image, progress) if mask is None: return None # Use the new create_editor_value function return create_editor_value(image, mask) def remove_objects( image: Image.Image | str | None, mask: Image.Image | str | None, user_data: dict, progress: gr.Progress = gr.Progress(), ) -> Image.Image | None: progress(0, desc="Preparing inputs...") if image is None or mask is None: return None if isinstance(mask, str): mask = Image.open(mask) if isinstance(image, str): image = Image.open(image) image_base64 = get_base64_from_image(image) mask_base64 = get_base64_from_image(mask) # Prepare inputs inputs: dict = { "image": f"data:image/png;base64,{image_base64}", "mask": f"data:image/png;base64,{mask_base64}", # "run_metatada": str( # { # "source": "HF", # "user": user_data, # } # ), } # Call ComfyDeploy API try: result = client.run.create( request={"deployment_id": CLEANER_DEPLOYMENT_ID, "inputs": inputs} ) if result and result.object: run_id: str = result.object.run_id progress(0, desc="Starting processing...") # Wait for the result while True: run_result = client.run.get(run_id=run_id) if not run_result.object: continue progress_value = ( run_result.object.progress if run_result.object.progress is not None else 0 ) status = ( run_result.object.live_status if run_result.object.live_status is not None else "Cold starting..." ) progress(progress_value, desc=f"Status: {status}") if run_result.object.status == "success": for output in run_result.object.outputs or []: if output.data and output.data.images: image_url: str = output.data.images[0].url # Download and return both the original and processed images response: requests.Response = requests.get(image_url) processed_image: Image.Image = Image.open( BytesIO(response.content) ) return processed_image return None elif run_result.object.status == "failed": logger.debug("Processing failed") return None time.sleep(1) # Wait for 1 second before checking the status again except Exception as e: logger.debug(f"Error: {e}") return None def resize_image(img: Image.Image, min_side_length: int = 768) -> Image.Image: if img.width <= min_side_length and img.height <= min_side_length: return img aspect_ratio = img.width / img.height if img.width < img.height: new_height = int(min_side_length / aspect_ratio) return img.resize((min_side_length, new_height)) new_width = int(min_side_length * aspect_ratio) return img.resize((new_width, min_side_length)) def get_profile(profile) -> dict: return { "username": profile.username, "profile": profile.profile, "name": profile.name, } async def run_removal( image_and_mask: EditorValue | None, progress: gr.Progress = gr.Progress(), profile: gr.OAuthProfile | None = None, ) -> tuple[Image.Image, Image.Image] | None: if not image_and_mask: gr.Info("Please upload an image and draw a mask") return None if profile is None: gr.Info("Please log in to process the image.") return None user_data = get_profile(profile) logger.debug("--------- RUN ----------") logger.debug(user_data) logger.debug("--------- RUN ----------") image_np = image_and_mask["background"] image_np = cast(np.ndarray, image_np) # If the image is empty, return None if np.sum(image_np) == 0: gr.Info("Please upload an image") return None alpha_channel = image_and_mask["layers"][0] alpha_channel = cast(np.ndarray, alpha_channel) mask_np = np.where(alpha_channel[:, :, 3] == 0, 0, 255).astype(np.uint8) # if mask_np is empty, return None if np.sum(mask_np) == 0: gr.Info("Please mark the areas you want to remove") return None mask = Image.fromarray(mask_np) mask = resize_image(mask) image = Image.fromarray(image_np) image = resize_image(image) output = remove_objects( image, # type: ignore mask, # type: ignore user_data, progress, ) if output is None: gr.Info("Processing failed") return None progress(100, desc="Processing completed") return image, output with gr.Blocks() as demo: gr.HTML("""
Upload an image and use the pencil tool (✏️ icon at the bottom) to mark the areas you want to remove.
For best results, include the shadows and reflections of the objects you want to remove. You can remove multiple objects at once. If you forget to mask some parts of your object, it's likely that the model will reconstruct them.
Finally, click on the "Run" button to process the image.
Wait for the processing to complete and compare the original and processed images using the slider.
⚠️ Note that the images are compressed to reduce the workloads of the demo.
Upload an image of the room you want to clean.
""") with gr.Row(equal_height=False): image_and_mask_auto = gr.ImageMask( label="Image and Mask", layers=False, show_fullscreen_button=False, sources=["upload"], show_download_button=False, interactive=True, height="full", width="full", brush=gr.Brush(default_size=75, colors=["#000000"], color_mode="fixed"), transforms=[], ) with gr.Column(): gr.HTML("""Click get mask to get automatic masking and edit it after manually if needed.
""") compute_mask_btn = gr.ClearButton( value="Get mask", variant="primary", size="lg", components=[image_and_mask_auto], ) compute_mask_btn.click( fn=lambda _: gr.update(interactive=False, value="Processing..."), inputs=[], outputs=[compute_mask_btn], api_name=False, ).then( fn=run_masking, inputs=[ input_image, ], outputs=[image_and_mask_auto], api_name=False, ).then( fn=lambda _: gr.update(interactive=True, value="Get mask"), inputs=[], outputs=[compute_mask_btn], api_name=False, ) # ------ REMOVAL with gr.Row(equal_height=False): image_slider = ImageSlider( label="Result", interactive=False, ) with gr.Column(): gr.HTML("""Click run to remove the objects from the image.
""") process_btn = gr.ClearButton( value="Run", variant="primary", size="lg", components=[image_slider], ) process_btn.click( fn=lambda _: gr.update(interactive=False, value="Processing..."), inputs=[], outputs=[process_btn], api_name=False, ).then( fn=run_removal, inputs=[ image_and_mask_auto, ], outputs=[image_slider], api_name=False, ).then( fn=lambda _: gr.update(interactive=True, value="Run"), inputs=[], outputs=[process_btn], api_name=False, ) if __name__ == "__main__": demo.launch( debug=False, share=False, show_api=False, )