app.py

import gradio as gr
import torch
import os
from diffusers import StableDiffusion3Pipeline
from safetensors.torch import load_file
from spaces import GPU  # Remove if not in HF Space

# 1. Define model ID and HF_TOKEN (at the VERY beginning)
model_id = "stabilityai/stable-diffusion-3.5-large"  # Or your preferred model ID
hf_token = os.getenv("HF_TOKEN")  # For private models (set in HF Space settings)

# 2. Initialize pipeline (to None initially)
pipeline = None

# 3. Load Stable Diffusion and LoRA (before Gradio)
try:
    if hf_token: # check if the token exists, if not, then do not pass the token
        pipeline = StableDiffusion3Pipeline.from_pretrained(
            model_id,
            torch_dtype=torch.float16,
            cache_dir="./model_cache"  # For caching
        )
    else:
        pipeline = StableDiffusion3Pipeline.from_pretrained(
            model_id,
            torch_dtype=torch.float16,
            cache_dir="./model_cache"  # For caching
        )

    lora_filename = "lora_trained_model.safetensors"  # EXACT filename of your LoRA
    lora_path = os.path.join("./", lora_filename)

    if os.path.exists(lora_path):
        lora_weights = load_file(lora_path)
        text_encoder = pipeline.text_encoder
        text_encoder.load_state_dict(lora_weights, strict=False)
        print(f"LoRA loaded successfully from: {lora_path}")
    else:
        print(f"Error: LoRA file not found at: {lora_path}")
        exit()  # Stop if LoRA is not found

    print("Stable Diffusion model loaded successfully!")

except Exception as e:
    print(f"Error loading model or LoRA: {e}")
    exit()  # Stop if model loading fails

# 4. Image generation function (now decorated)
@GPU(duration=65)  # ONLY if in a HF Space, remove if not
def generate_image(prompt):
    global pipeline
    if pipeline is None:  # Should not happen, but good to check
        return "Error: Model not loaded!"

    try:
        image = pipeline(prompt).images[0]  # Access the first image from the list
        return image
    except Exception as e:
        return f"Error generating image: {e}"

# 5. Gradio interface
with gr.Blocks() as demo:
    prompt_input = gr.Textbox(label="Prompt")
    image_output = gr.Image(label="Generated Image")
    generate_button = gr.Button("Generate")

    generate_button.click(
        fn=generate_image,
        inputs=prompt_input,
        outputs=image_output,
    )

demo.launch()