app.py

import gradio as gr
import torch
from diffusers import StableDiffusionPipeline
import os

print("--- FINAL VERSION: ADJUSTING LORA SCALE ---")

# --- Step 1: Define constants ---
BASE_MODEL_ID = "runwayml/stable-diffusion-v1-5"
LORA_FILENAME = "MyStickmanProject-10.safetensors"

print(f"Base model ID: {BASE_MODEL_ID}")
print(f"LoRA file: {LORA_FILENAME}")

# --- Step 2: Load the full base model pipeline ---
print("Loading base model pipeline...")
pipe = StableDiffusionPipeline.from_pretrained(
    BASE_MODEL_ID,
    use_safetensors=True
)
print("Base model loaded successfully.")

# --- Step 3: Load and apply the LoRA weights ---
# This is the standard way to load LoRA weights.
print("Loading LoRA weights...")
pipe.load_lora_weights(LORA_FILENAME)
print("LoRA weights loaded.")


# --- Step 4: Move to CPU and optimize ---
print("Moving pipeline to CPU...")
pipe.to("cpu")
print("Enabling attention slicing...")
pipe.enable_attention_slicing()
print("--- MODEL SETUP COMPLETE ---")

# --- Step 5: Define the core generation function ---
def generate_image(prompt, negative_prompt, guidance_scale, num_steps, lora_scale):
    print(f"Generating image with prompt: {prompt}")
    
    # This is the KEY CHANGE: We are now controlling the LoRA strength.
    # lora_scale=1.0 means full strength. lora_scale=0.7 means 70% strength.
    image = pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        num_inference_steps=int(num_steps),
        guidance_scale=float(guidance_scale),
        cross_attention_kwargs={"scale": lora_scale}
    ).images[0]
    
    print("Image generation complete.")
    return image

# --- Step 6: Create the Gradio user interface ---
with gr.Blocks() as iface:
    gr.Markdown("# My Custom Stickman LoRA Demo")
    gr.Markdown("Enter a prompt to generate an image in my unique style.")
    
    with gr.Row():
        with gr.Column(scale=70):
            prompt_input = gr.Textbox(label="Prompt", placeholder="A stickman wearing a fedora hat, holding a magnifying glass")
            negative_prompt_input = gr.Textbox(label="Negative Prompt", value="(worst quality, low quality:1.4), blurry, noisy, grainy, 3d, realistic, photo")
            
            # We add a new slider to control the LoRA strength!
            lora_slider = gr.Slider(minimum=0, maximum=1.5, step=0.05, value=0.75, label="LoRA Strength (Style Intensity)")

            with gr.Row():
                guidance_scale_slider = gr.Slider(minimum=1, maximum=20, step=0.5, value=7.5, label="Guidance Scale")
                steps_slider = gr.Slider(minimum=10, maximum=100, step=1, value=25, label="Inference Steps")
        
        with gr.Column(scale=30):
            image_output = gr.Image(label="Generated Image")

    submit_button = gr.Button("Generate", variant="primary")
    submit_button.click(
        fn=generate_image,
        inputs=[prompt_input, negative_prompt_input, guidance_scale_slider, steps_slider, lora_slider],
        outputs=image_output
    )

# --- Step 7: Launch the application ---
print("Launching Gradio app...")
iface.launch()