import gradio as gr
import spaces
import random

import torch
from diffusers import FluxPipeline
from huggingface_hub.utils import RepositoryNotFoundError

pipeline = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=torch.float16).to("cuda")

@spaces.GPU(duration=70)
def generate(prompt, negative_prompt, width, height, sample_steps, lora_id):
    try:
        pipeline.load_lora_weights(lora_id)
    except RepositoryNotFoundError:
        raise ValueError(f"Recieved invalid FLUX LoRA.")
    
    return pipeline(prompt=f"{prompt}\n(NOT {negative_prompt}:2)", width=width, height=height, num_inference_steps=sample_steps, generator=torch.Generator("cpu").manual_seed(random.randint(42, 69)), guidance_scale=7).images[0]

with gr.Blocks() as interface:
        with gr.Column():
            with gr.Row():
                with gr.Column():
                    prompt = gr.Textbox(label="Prompt", info="What do you want?", value="Keanu Reeves holding a neon sign reading 'Hello, world!', 32k HDR, paparazzi", lines=4, interactive=True)
                    negative_prompt = gr.Textbox(label="Negative Prompt", info="What do you want to exclude from the image?", value="ugly, low quality", lines=4, interactive=True)
                with gr.Column():
                    generate_button = gr.Button("Generate")
                    output = gr.Image()
            with gr.Row():
                with gr.Accordion(label="Advanced Settings", open=False):
                    with gr.Row():
                        with gr.Column():
                            width = gr.Slider(label="Width", info="The width in pixels of the generated image.", value=512, minimum=128, maximum=4096, step=64, interactive=True)
                            height = gr.Slider(label="Height", info="The height in pixels of the generated image.", value=512, minimum=128, maximum=4096, step=64, interactive=True)
                        with gr.Column():
                            sampling_steps = gr.Slider(label="Sampling Steps", info="The number of denoising steps.", value=20, minimum=4, maximum=50, step=1, interactive=True)
                            lora_id = gr.Textbox(label="Adapter Repository", info="ID of the FLUX LoRA", value="pepper13/fluxfw")
        
        generate_button.click(fn=generate, inputs=[prompt, negative_prompt, width, height, sampling_steps, lora_id], outputs=[output])

if __name__ == "__main__":
    interface.launch()