import spaces import gradio as gr import numpy as np import torch from PIL import Image from diffusers import DDPMScheduler, StableDiffusionPipeline, DDIMScheduler, UNet2DConditionModel from diffusers import StableDiffusionInstructPix2PixPipeline, LCMScheduler # InstructPix2Pix with LCM specified scheduler pipe = StableDiffusionInstructPix2PixPipeline.from_pretrained( "timbrooks/instruct-pix2pix", torch_dtype=torch.float16 ) pipe = pipe.to("cuda") pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config) # Adapt the InstructPix2Pix model using the LoRA parameters adapter_id = "latent-consistency/lcm-lora-sdv1-5" pipe.load_lora_weights(adapter_id) pipe.to('cuda') MAX_SEED = np.iinfo(np.int32).max MAX_IMAGE_SIZE = 1024 @spaces.GPU(duration=30) def infer(image, edit_instruction, guidance_scale, n_steps): image = Image.fromarray(image).resize((512, 512)) image = pipe(prompt=edit_instruction, image=image, num_inference_steps=n_steps, guidance_scale=guidance_scale, ).images[0] return image css=""" #col-container { margin: 0 auto; max-width: 1024px; } """ if torch.cuda.is_available(): power_device = "GPU" else: power_device = "CPU" with gr.Blocks(css=css) as demo: with gr.Column(elem_id="col-container"): gr.Markdown( f""" # ⚡ Instruct-pix2pix with Consistency Distillation⚡ Currently running on {power_device} """ ) gr.Markdown( "If you enjoy the space, feel free to give a ⭐ to the Github Repo. [![GitHub Stars](https://img.shields.io/github/stars/quickjkee/instruct-pix2pix-distill?style=social)](https://github.com/quickjkee/instruct-pix2pix-distill)" ) with gr.Row(): edit_instruction = gr.Text( label="Edit instruction", max_lines=1, placeholder="Enter your prompt", ) with gr.Row(): with gr.Column(): image = gr.Image(label="Input image", height=512, width=512, show_label=False) with gr.Column(): result = gr.Image(label="Result", height=512, width=512, show_label=False) with gr.Accordion("Advanced Settings", open=True): with gr.Row(): guidance_scale = gr.Slider( label="guidance scale", minimum=1.0, maximum=5.0, step=1.0, value=2.0, ) n_steps = gr.Slider( label="inference steps", minimum=1.0, maximum=10.0, step=1.0, value=4.0, ) with gr.Row(): run_button = gr.Button("Edit", scale=0) with gr.Row(): examples = [ [ "examples/orig_3.jpg", #input_image "a photo of a basket of apples", #src_prompt "a photo of a basket of oranges", #tgt_prompt 20, #guidance_scale 0.6, #tau 0.4, #crs 0.6, #srs 1, #amplify factor 'oranges', # amplify word '', #orig blend 'oranges', #edited blend False #replacement ], [ "examples/orig_3.jpg", #input_image "a photo of a basket of apples", #src_prompt "a photo of a basket of puppies", #tgt_prompt 20, #guidance_scale 0.6, #tau 0.4, #crs 0.1, #srs 2, #amplify factor 'puppies', # amplify word '', #orig blend 'puppies', #edited blend True #replacement ], [ "examples/orig_3.jpg", #input_image "a photo of a basket of apples", #src_prompt "a photo of a basket of apples under snowfall", #tgt_prompt 20, #guidance_scale 0.6, #tau 0.4, #crs 0.4, #srs 30, #amplify factor 'snowfall', # amplify word '', #orig blend 'snowfall', #edited blend False #replacement ], [ "examples/orig_1.jpg", #input_image "a photo of an owl", #src_prompt "a photo of an yellow owl", #tgt_prompt 20, #guidance_scale 0.6, #tau 0.9, #crs 0.9, #srs 20, #amplify factor 'yellow', # amplify word 'owl', #orig blend 'yellow', #edited blend False #replacement ], [ "examples/orig_1.jpg", #input_image "a photo of an owl", #src_prompt "an anime-style painting of an owl", #tgt_prompt 20, #guidance_scale 0.8, #tau 0.6, #crs 0.3, #srs 10, #amplify factor 'anime-style', # amplify word 'painting', #orig blend 'anime-style', #edited blend False #replacement ], [ "examples/orig_1.jpg", #input_image "a photo of an owl", #src_prompt "a photo of an owl underwater with many fishes nearby", #tgt_prompt 20, #guidance_scale 0.8, #tau 0.4, #crs 0.4, #srs 18, #amplify factor 'fishes', # amplify word '', #orig blend 'fishes', #edited blend False #replacement ], [ "examples/orig_2.jpg", #input_image "a photograph of a teddy bear sitting on a wall", #src_prompt "a photograph of a teddy bear sitting on a wall surrounded by roses", #tgt_prompt 20, #guidance_scale 0.6, #tau 0.4, #crs 0.1, #srs 25, #amplify factor 'roses', # amplify word '', #orig blend 'roses', #edited blend False #replacement ], [ "examples/orig_2.jpg", #input_image "a photograph of a teddy bear sitting on a wall", #src_prompt "a photograph of a wooden bear sitting on a wall", #tgt_prompt 20, #guidance_scale 0.8, #tau 0.5, #crs 0.5, #srs 14, #amplify factor 'wooden', # amplify word '', #orig blend 'wooden', #edited blend True #replacement ], [ "examples/orig_2.jpg", #input_image "a photograph of a teddy bear sitting on a wall", #src_prompt "a photograph of a teddy rabbit sitting on a wall", #tgt_prompt 20, #guidance_scale 0.8, #tau 0.4, #crs 0.4, #srs 3, #amplify factor 'rabbit', # amplify word '', #orig blend 'rabbit', #edited blend True #replacement ], ] #gr.Examples( # examples = examples, # inputs =[input_image, input_prompt, prompt, # guidance_scale, tau, crs, srs, amplify_factor, amplify_word, # blend_orig, blend_edited, is_replacement], # outputs=[ # result # ], # fn=infer, cache_examples=True #) run_button.click( fn = infer, inputs=[image, edit_instruction, guidance_scale, n_steps], outputs = [result] ) demo.queue().launch()