import gradio as gr from PIL import Image import base64 import io import glob import cv2 import numpy as np import torch from controlnet_aux import HEDdetector from diffusers import StableDiffusionControlNetPipeline, ControlNetModel, UniPCMultistepScheduler def predict(sketch, description): # Convert sketch to PIL image sketch_pil = Image.fromarray(sketch) hed = HEDdetector.from_pretrained('lllyasviel/Annotators') image = hed(sketch_pil, scribble=True) model_id = "runwayml/stable-diffusion-v1-5" controlnet_id = "lllyasviel/sd-controlnet-scribble" # Load ControlNet model controlnet = ControlNetModel.from_pretrained(controlnet_id) # Create pipeline with ControlNet model pipe = StableDiffusionControlNetPipeline.from_pretrained(model_id, controlnet=controlnet) # Use improved scheduler pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config) # Enable smart CPU offloading and memory efficient attention # pipe.enable_model_cpu_offload() # pipe.enable_xformers_memory_efficient_attention() # Move pipeline to GPU # pipe = pipe.to("cuda") result = pipe(description, image, num_inference_steps=10).images[0] return result with gr.Blocks() as iface: # Define sketchpad with custom size and stroke width sketchpad = gr.Sketchpad(shape=(400, 300), brush_radius=5, label="Sketchpad- Draw something") txt= gr.Textbox(lines=3, label="Description - Describe your sketch with style") im = gr.Image(label="Output Image", interactive=False) button = gr.Button(value="Submit") button.click(predict, inputs=[sketchpad, txt], outputs=im) flag= gr.CSVLogger() flag.setup([sketchpad, txt, im], "flagged_data_points") button_flag = gr.Button(value="Flag") button_flag.click(lambda *args: flag.flag(args), [sketchpad, txt, im], None, preprocess=False) # iface = gr.Interface(fn=predict, inputs=[sketchpad, "text"], outputs=im, live=False, title="Sketch2Image") ## get all the file path from flagged/sketch folder into a list sketch_path = glob.glob("flagged/sketch/*.png") # gr.Examples(examples = list(map(lambda x: [x ,"draw in the style of crayon by kids"], sketch_path)), inputs=[sketchpad,txt], outputs=im, fn=predict, cache_examples=True) iface.launch()