from rembg import remove, new_session import gradio as gr import numpy as np import random from typing import Optional # import spaces #[uncomment to use ZeroGPU] from diffusers import DiffusionPipeline, AutoencoderTiny from diffusers import ( DPMSolverMultistepScheduler, DPMSolverSinglestepScheduler, KDPM2DiscreteScheduler, KDPM2AncestralDiscreteScheduler, EulerDiscreteScheduler, EulerAncestralDiscreteScheduler, HeunDiscreteScheduler, LMSDiscreteScheduler, DDIMScheduler ) from peft import PeftModel 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 lora_map={"xenomirant/frieren_sd-1-4-lora": "CompVis/stable-diffusion-v1-4", "xenomirant/frieren_waifu-dd-lora": "hakurei/waifu-diffusion"} if torch.cuda.is_available(): torch_dtype = torch.float16 else: torch_dtype = torch.float32 MAX_SEED = np.iinfo(np.int32).max MAX_IMAGE_SIZE = 512 # @spaces.GPU #[uncomment to use ZeroGPU] def infer( model_id, prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, lora_scale: Optional[float] = .5, scheduler: Optional[str] = None, use_tiny_vae: Optional[bool] = False, remove_background: Optional[bool] = True, progress=gr.Progress(track_tqdm=True), ): if "lora" in model_id: pipe = DiffusionPipeline.from_pretrained(lora_map[model_id], torch_dtype=torch_dtype) pipe.unet = PeftModel.from_pretrained(pipe.unet, model_id, subfolder="unet", ) pipe.text_encoder = PeftModel.from_pretrained(pipe.text_encoder, model_id, subfolder="text_encoder", ) else: pipe = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch_dtype) if use_tiny_vae: pipe.vae = AutoencoderTiny.from_pretrained("madebyollin/taesd", torch_dtype=torch.float16) match scheduler: case None: pass case "DPMSolverMultistepScheduler": if DPMSolverMultistepScheduler in pipe.scheduler.compatibles: scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config) pipe.scheduler = scheduler case "DPMSolverSinglestepScheduler": if DPMSolverSinglestepScheduler in pipe.scheduler.compatibles: scheduler = DPMSolverSinglestepScheduler.from_config(pipe.scheduler.config) pipe.scheduler = scheduler case "KDPM2DiscreteScheduler": if KDPM2DiscreteScheduler in pipe.scheduler.compatibles: scheduler = KDPM2DiscreteScheduler.from_config(pipe.scheduler.config) pipe.scheduler = scheduler case "KDPM2AncestralDiscreteScheduler": if KDPM2AncestralDiscreteScheduler in pipe.scheduler.compatibles: scheduler = KDPM2AncestralDiscreteScheduler.from_config(pipe.scheduler.config) pipe.scheduler = scheduler case "EulerDiscreteScheduler": if EulerDiscreteScheduler in pipe.scheduler.compatibles: scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config) pipe.scheduler = scheduler case "EulerAncestralDiscreteScheduler": if EulerAncestralDiscreteScheduler in pipe.scheduler.compatibles: scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config) pipe.scheduler = scheduler case "HeunDiscreteScheduler": if HeunDiscreteScheduler in pipe.scheduler.compatibles: scheduler = HeunDiscreteScheduler.from_config(pipe.scheduler.config) pipe.scheduler = scheduler case "LMSDiscreteScheduler": if LMSDiscreteScheduler in pipe.scheduler.compatibles: scheduler = LMSDiscreteScheduler.from_config(pipe.scheduler.config) pipe.scheduler = scheduler case "DDIMScheduler": if DDIMScheduler in pipe.scheduler.compatibles: scheduler = DDIMScheduler.from_config(pipe.scheduler.config, timestep_spacing="trailing", ) pipe.scheduler = scheduler pipe = pipe.to(device) if randomize_seed: seed = random.randint(0, MAX_SEED) generator = torch.Generator().manual_seed(seed) image = pipe( prompt=prompt, negative_prompt=negative_prompt, guidance_scale=guidance_scale, num_inference_steps=num_inference_steps, width=width, height=height, generator=generator, cross_attention_kwargs={"scale": lora_scale} if "lora" in model_id else {} ).images[0] if remove_background: session = new_session("isnet-anime",) image = remove(image, alpha_matting=True, alpha_matting_erode_size=15, session=session ) return image, seed examples = [ "onnxwitchwitch, an elf with a questioned expression, castle at her back, muted colors, detailed, 8k", "onnxwitchwitch, an elf with a huge cloak with a feeling of horror, golden cave, detailed, ", "onnxwitchwitch, an elf in a dress sitting on a pile of flowers, bright colors, sunny, 4k", ] css = """ #col-container { margin: 0 auto; max-width: 640px; } """ with gr.Blocks(css=css) as demo: with gr.Column(elem_id="col-container"): gr.Markdown(" # Text-to-Image Gradio Template") model_id = gr.Dropdown( ["xenomirant/frieren_sd-1-4-lora", "xenomirant/frieren_waifu-dd-lora", "CompVis/stable-diffusion-v1-4", "hakurei/waifu-diffusion"], label="Image-to-text model", visible=True, ) 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") use_tiny_vae = gr.Checkbox( label="Use tiny VAE for faster inference?", ) remove_background = gr.Checkbox( label="Remove background?", ) 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=True, ) seed = gr.Slider( label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=42, ) randomize_seed = gr.Checkbox(label="Randomize seed", value=True) scheduler = gr.Dropdown( [None, "DPMSolverMultistepScheduler", "DPMSolverSinglestepScheduler", "KDPM2DiscreteScheduler", "KDPM2AncestralDiscreteScheduler", "EulerDiscreteScheduler", "EulerAncestralDiscreteScheduler", "HeunDiscreteScheduler", "LMSDiscreteScheduler", "DDIMScheduler"], label="Scheduler", visible=True ) with gr.Row(): width = gr.Slider( label="Width", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=512, # Replace with defaults that work for your model ) height = gr.Slider( label="Height", minimum=256, maximum=MAX_IMAGE_SIZE, step=32, value=512, # 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=7.0, # 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=20, # Replace with defaults that work for your model ) lora_scale = gr.Slider( label="LoRA weight during generation", minimum=0.0, maximum=1.0, step=0.05, value=0.75, # Replace with defaults that work for your model ) gr.Examples(examples=examples, inputs=[prompt]) gr.on( triggers=[run_button.click, prompt.submit], fn=infer, inputs=[ model_id, prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, lora_scale, scheduler, use_tiny_vae, remove_background ], outputs=[result, seed], ) if __name__ == "__main__": demo.launch()