import gradio as gr
import json
import logging
import torch
from PIL import Image
from diffusers import DiffusionPipeline, EulerDiscreteScheduler, DPMSolverMultistepScheduler
import spaces

# Load LoRAs from JSON file
with open('loras.json', 'r') as f:
    loras = json.load(f)

# Initialize the base model
base_model = "stabilityai/stable-diffusion-xl-base-1.0"
pipe = DiffusionPipeline.from_pretrained(base_model, torch_dtype=torch.float16)
pipe.to("cuda")

def update_selection(evt: gr.SelectData):
    selected_lora = loras[evt.index]
    new_placeholder = f"Type a prompt for {selected_lora['title']}"
    lora_repo = selected_lora["repo"]
    updated_text = f"### Selected: [{lora_repo}](https://huggingface.co/{lora_repo}) ✨"
    return (
        gr.update(placeholder=new_placeholder),
        updated_text,
        evt.index
    )

@spaces.GPU
def run_lora(prompt, negative_prompt, cfg_scale, steps, selected_index, scheduler):
    if selected_index is None:
        raise gr.Error("You must select a LoRA before proceeding.")

    selected_lora = loras[selected_index]
    lora_path = selected_lora["repo"]
    trigger_word = selected_lora["trigger_word"]

    # Load LoRA weights
    pipe.load_lora_weights(lora_path)

    # Set scheduler
    if scheduler == "Euler":
        pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config)
    elif scheduler == "DPM++ 2M":
        pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)

    # Generate image
    image = pipe(
        prompt=f"{prompt} {trigger_word}",
        negative_prompt=negative_prompt,
        num_inference_steps=steps,
        guidance_scale=cfg_scale,
    ).images[0]

    # Unload LoRA weights
    pipe.unload_lora_weights()

    return image

with gr.Blocks(css="custom.css") as app:
    gr.Markdown("# artificialguybr LoRA portfolio")
    gr.Markdown(
        "### This is my portfolio. Follow me on Twitter [@artificialguybr](https://twitter.com/artificialguybr).\n"
        "**Note**: Generation quality may vary. For best results, adjust the parameters.\n"
        "Special thanks to Hugging Face for their Diffusers library and Spaces platform."
    )

    selected_index = gr.State(None)

    with gr.Row():
        gallery = gr.Gallery(
            [(item["image"], item["title"]) for item in loras],
            label="LoRA Gallery",
            allow_preview=False,
            columns=3
        )

        with gr.Column():
            prompt_title = gr.Markdown("### Click on a LoRA in the gallery to select it")
            selected_info = gr.Markdown("")
            prompt = gr.Textbox(label="Prompt", lines=3, placeholder="Type a prompt after selecting a LoRA")
            negative_prompt = gr.Textbox(label="Negative Prompt", lines=2, value="low quality, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry")
            
            with gr.Row():
                cfg_scale = gr.Slider(label="CFG Scale", minimum=1, maximum=20, step=0.5, value=7.5)
                steps = gr.Slider(label="Steps", minimum=1, maximum=100, step=1, value=30)
            
            scheduler = gr.Dropdown(label="Scheduler", choices=["Euler", "DPM++ 2M"], value="Euler")
            
            generate_button = gr.Button("Generate")
            result = gr.Image(label="Generated Image")

    gallery.select(update_selection, outputs=[prompt, selected_info, selected_index])
    
    generate_button.click(
        fn=run_lora,
        inputs=[prompt, negative_prompt, cfg_scale, steps, selected_index, scheduler],
        outputs=[result]
    )

app.queue()
app.launch()