import gradio as gr from PIL import Image from io import BytesIO import torch import os from diffusers import DiffusionPipeline, DDIMScheduler, UNet2DConditionModel MY_SECRET_TOKEN=os.environ.get('HF_TOKEN_SD') has_cuda = torch.cuda.is_available() device = "cuda" pipe = DiffusionPipeline.from_pretrained( "CompVis/stable-diffusion-v1-4", safety_checker=None, custom_pipeline="imagic_stable_diffusion", scheduler = DDIMScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", clip_sample=False, set_alpha_to_one=False) ).to(device) generator = th.Generator("cuda").manual_seed(0) def infer(prompt, init_image): init_image = Image.open(init_image).convert("RGB") init_image = init_image.resize((128, 128)) res = pipe.train( prompt, init_image, guidance_scale=7.5, num_inference_steps=50, generator=generator, text_embedding_optimization_steps=100, model_fine_tuning_optimization_steps=500) #with torch.no_grad(): # torch.cuda.empty_cache() #res = pipe(alpha=1) #return res.images[0] return 'trained success' title = """
Text-Based Real Image Editing with Diffusion Models
This pipeline aims to implement this paper to Stable Diffusion, allowing for real-world image editing.