app.py

import gradio as gr
import spaces
import random
import torch
from huggingface_hub import snapshot_download
from kolors.pipelines.pipeline_stable_diffusion_xl_chatglm_256_inpainting import StableDiffusionXLInpaintPipeline
from kolors.models.modeling_chatglm import ChatGLMModel
from kolors.models.tokenization_chatglm import ChatGLMTokenizer
from diffusers import AutoencoderKL, EulerDiscreteScheduler, UNet2DConditionModel
from groundingdino.util.inference import load_model, predict
from segment_anything import SamAutomaticMaskGenerator
from PIL import Image
import numpy as np
import os

# Download model checkpoints
device = "cuda"
ckpt_dir = snapshot_download(repo_id="Kwai-Kolors/Kolors-Inpainting")

# Inpainting setup
text_encoder = ChatGLMModel.from_pretrained(f'{ckpt_dir}/text_encoder',torch_dtype=torch.float16).half().to(device)
tokenizer = ChatGLMTokenizer.from_pretrained(f'{ckpt_dir}/text_encoder')
vae = AutoencoderKL.from_pretrained(f"{ckpt_dir}/vae", revision=None).half().to(device)
scheduler = EulerDiscreteScheduler.from_pretrained(f"{ckpt_dir}/scheduler")
unet = UNet2DConditionModel.from_pretrained(f"{ckpt_dir}/unet", revision=None).half().to(device)

pipe = StableDiffusionXLInpaintPipeline(
    vae=vae,
    text_encoder=text_encoder,
    tokenizer=tokenizer,
    unet=unet,
    scheduler=scheduler
)

pipe.to(device)
pipe.enable_attention_slicing()

# GroundingDINO and SAM setup
model_dino = load_model("path/to/groundingdino/config.yaml", "path/to/groundingdino/model.pth")
sam = SamAutomaticMaskGenerator(model_type="vit_h", checkpoint="model/sam_vit_h_4b8939.pth")

# Constants
MAX_SEED = np.iinfo(np.int32).max

def generate_mask(image: Image):
    boxes, logits, phrases = predict(model_dino, image, "prompt")  # Provide the proper prompt for detection
    masks = sam.generate(image)
    mask = masks[0]["segmentation"]  # Use the first detected mask as an example
    return Image.fromarray(mask)

@spaces.GPU
def infer(prompt, image, negative_prompt, seed, randomize_seed, guidance_scale, num_inference_steps):
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    
    # Generate mask using GroundingDINO + SAM
    mask_image = generate_mask(image)
    
    generator = torch.Generator().manual_seed(seed)
    result = pipe(
        prompt=prompt,
        image=image,
        mask_image=mask_image,
        height=image.height,
        width=image.width,
        guidance_scale=guidance_scale,
        generator=generator,
        num_inference_steps=num_inference_steps,
        negative_prompt=negative_prompt,
        num_images_per_prompt=1,
        strength=0.999
    ).images[0]
    
    return result

css="""
#col-left {
    margin: 0 auto;
    max-width: 600px;
}
#col-right {
    margin: 0 auto;
    max-width: 700px;
}
"""

def load_description(fp):
    with open(fp, 'r', encoding='utf-8') as f:
        content = f.read()
    return content

with gr.Blocks(css=css) as Kolors:
    gr.HTML(load_description("assets/title.md"))
    
    with gr.Row():
        with gr.Column(elem_id="col-left"):
            prompt = gr.Textbox(label="Prompt", placeholder="Enter your prompt", lines=2)
            image = gr.ImageEditor(label="Image", type="pil", image_mode='RGB')
            
            with gr.Accordion("Advanced Settings", open=False):
                negative_prompt = gr.Textbox(label="Negative prompt", value="low quality, bad anatomy")
                seed = gr.Slider(label="Seed", minimum=0, maximum=MAX_SEED, step=1, value=0)
                randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
                guidance_scale = gr.Slider(label="Guidance scale", minimum=0.0, maximum=10.0, step=0.1, value=6.0)
                num_inference_steps = gr.Slider(label="Number of inference steps", minimum=10, maximum=50, step=1, value=25)
            
            run_button = gr.Button("Run")
        
        with gr.Column(elem_id="col-right"):
            result = gr.Image(label="Result", show_label=False)
    
    run_button.click(
        fn=infer,
        inputs=[prompt, image, negative_prompt, seed, randomize_seed, guidance_scale, num_inference_steps],
        outputs=[result]
    )

Kolors.queue().launch(debug=True)