app.py

import gradio as gr
import torch
from PIL import Image
import model_loader
import pipeline
from transformers import CLIPTokenizer

DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Using device: {DEVICE}")

# Load tokenizer and model
model_file = "./data/v1-5-pruned-emaonly.ckpt"
tokenizer = CLIPTokenizer("./data/vocab.json", merges_file="./data/merges.txt")
models = model_loader.preload_models_from_standard_weights(model_file, DEVICE)

def generate_image(prompt, uncond_prompt, cfg_scale, sampler, num_steps, seed, image, strength, width, height):
    input_image = Image.open(image) if image else None
    output_image = pipeline.generate(
        prompt=prompt,
        uncond_prompt=uncond_prompt,
        input_image=input_image,
        strength=strength,
        do_cfg=True,
        cfg_scale=cfg_scale,
        sampler_name=sampler,
        n_inference_steps=num_steps,
        seed=int(seed) if seed else None,
        models=models,
        device=DEVICE,
        idle_device="cpu",
        tokenizer=tokenizer,
    )
    return Image.fromarray(output_image)

with gr.Blocks() as demo:
    gr.Markdown("# Text-to-Image Gradio Interface")
    prompt = gr.Textbox(label="Prompt", value="Astronaut in a jungle, cold color palette, muted colors, detailed, 8k")
    generate_btn = gr.Button("Run")
    output_image = gr.Image(label="Generated Image")
    
    with gr.Accordion("Advanced Settings", open=False):
        seed = gr.Number(value=42, label="Seed", interactive=True)
        randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
        cfg_scale = gr.Slider(1, 14, value=8, step=0.5, label="CFG Scale")
        sampler = gr.Dropdown(["ddpm", "ddim", "plms"], value="ddpm", label="Sampler")
        num_steps = gr.Slider(1, 100, value=50, step=1, label="Number of inference steps")
        image = gr.File(label="Upload Image (Optional)")
        strength = gr.Slider(0, 1, value=0.75, step=0.05, label="Strength (for Image-to-Image)")
    
    generate_btn.click(
        generate_image, 
        inputs=[prompt, gr.Textbox(value="", visible=False), cfg_scale, sampler, num_steps, seed, image, strength],
        outputs=output_image
    )

demo.launch()