app.py

import torch
from diffusers import StableDiffusionXLPipeline, UNet2DConditionModel
from transformers.utils.hub import cached_file
from safetensors.torch import load_file
import gradio as gr

# --- Configuration ---
base_model = "stabilityai/stable-diffusion-xl-base-1.0"
lightning_repo = "ByteDance/SDXL-Lightning"
checkpoint_filename = "sdxl_lightning_2step_unet.safetensors"
device = "cpu"  # Force CPU

# --- Download and Load Lightning UNet ---
print("Downloading Lightning checkpoint...")
ckpt_path = cached_file(lightning_repo, checkpoint_filename, cache_dir=".cache")

print("Loading UNet (CPU)...")
unet = UNet2DConditionModel.from_config(base_model, subfolder="unet")
unet.load_state_dict(load_file(ckpt_path))
unet.to(device)
unet.eval()

# --- Load Pipeline without FP16 ---
print("Loading Stable Diffusion XL Pipeline...")
pipe = StableDiffusionXLPipeline.from_pretrained(
    base_model,
    unet=unet
)
pipe.to(device)

# --- Generation Function ---
def generate_image(prompt):
    if not prompt:
        return "Prompt cannot be empty!"
    image = pipe(prompt, num_inference_steps=2, guidance_scale=0).images[0]
    return image

# --- Example Prompts ---
examples = [
    ["A futuristic city skyline at sunset, ultra-detailed, sci-fi style"],
    ["An astronaut riding a horse on Mars, digital art"],
    ["A serene forest landscape with glowing mushrooms, fantasy art"],
    ["Cyberpunk samurai under neon lights, raining scene"],
    ["A cute robot chef in a cozy kitchen, Pixar style"]
]

# --- Gradio UI ---
demo = gr.Interface(
    fn=generate_image,
    inputs=gr.Textbox(label="Enter your prompt", placeholder="e.g., A castle floating in the clouds"),
    outputs=gr.Image(type="pil"),
    title="SDXL-Lightning (2-Step, CPU) Image Generator",
    description="Fast image generation using ByteDance's SDXL-Lightning 2-step model on CPU (no optimization).",
    examples=examples,
    cache_examples=False
)

# --- Launch Interface ---
demo.launch()