app.py

import spaces
import gradio as gr
import torch
from diffusers import LCMScheduler, AutoPipelineForText2Image
from diffusers import AutoPipelineForInpainting, LCMScheduler
from diffusers import DiffusionPipeline, LCMScheduler
from PIL import Image, ImageEnhance
import io

pipe = DiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    variant="fp16",
    torch_dtype=torch.float32
).to("cuda")

# set scheduler
pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)

# Load and fuse lcm lora
pipe.load_lora_weights("latent-consistency/lcm-lora-sdxl", adapter_name="lcm")
pipe.load_lora_weights("Pclanglais/wiki-model", weight_name="pytorch_lora_weights.safetensors", adapter_name="mickey")

# Combine LoRAs
pipe.set_adapters(["lcm", "mickey"], adapter_weights=[1.0, 1.0])
pipe.fuse_lora()


@spaces.GPU
def generate_image(prompt, num_inference_steps, guidance_scale):
    model_id = "stabilityai/stable-diffusion-xl-base-1.0"
    adapter_id = "latent-consistency/lcm-lora-sdxl"

    pipe = AutoPipelineForText2Image.from_pretrained(model_id, torch_dtype=torch.float32, variant="fp16")
    pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
    pipe.to("cuda")

    # Load and fuse lcm lora
    pipe.load_lora_weights(adapter_id)
    pipe.fuse_lora()

    # Generate the image
    image = pipe(prompt=prompt, num_inference_steps=num_inference_steps, guidance_scale=guidance_scale).images[0]
  
    return image

def inpaint_image(prompt, init_image, mask_image, num_inference_steps, guidance_scale):
    pipe = AutoPipelineForInpainting.from_pretrained(
        "diffusers/stable-diffusion-xl-1.0-inpainting-0.1",
        torch_dtype=torch.float32,
        variant="fp16",
    ).to("cuda")
    pipe.scheduler = LCMScheduler.from_config(pipe.scheduler.config)
    pipe.load_lora_weights("latent-consistency/lcm-lora-sdxl")
    pipe.fuse_lora()

    if init_image is not None:
        init_image_path = init_image.name  # Get the file path
        init_image = Image.open(init_image_path).resize((1024, 1024))
    else:
        raise ValueError("Initial image not provided or invalid")

    if mask_image is not None:
        mask_image_path = mask_image.name  # Get the file path
        mask_image = Image.open(mask_image_path).resize((1024, 1024))
    else:
        raise ValueError("Mask image not provided or invalid")

    # Generate the inpainted image
    generator = torch.manual_seed(42)
    image = pipe(
        prompt=prompt,
        image=init_image,
        mask_image=mask_image,
        generator=generator,
        num_inference_steps=num_inference_steps,
        guidance_scale=guidance_scale,
    ).images[0]

    return image

def generate_image_with_adapter(pipe, prompt, num_inference_steps, guidance_scale):
    generator = torch.manual_seed(0)
    # Generate the image
    image = pipe(prompt=prompt, num_inference_steps=num_inference_steps, guidance_scale=guidance_scale, generator=generator).images[0]
    return image


with gr.Blocks(gr.themes.Soft()) as demo:
    with gr.Row():
        image_output = gr.Image(label="Generated Image")
    with gr.Row():
        with gr.Accordion(label="Wiki-Mouse Image Generation"):
            adapter_prompt_input = gr.Textbox(label="Prompt", placeholder="papercut, a cute fox")
            adapter_steps_input = gr.Slider(minimum=1, maximum=10, label="Inference Steps", value=4)
            adapter_guidance_input = gr.Slider(minimum=0, maximum=2, label="Guidance Scale", value=1)
            adapter_generate_button = gr.Button("Generate Image with Adapter")

    adapter_generate_button.click(
        generate_image_with_adapter,
        inputs=[pipe, adapter_prompt_input, adapter_steps_input, adapter_guidance_input],
        outputs=image_output
    )

demo.launch()