app.py

import spaces
import torch
import gradio as gr
from PIL import Image
import os

from diffusers import (
    DiffusionPipeline,
    AutoencoderKL,
    ControlNetModel,
    StableDiffusionControlNetPipeline,
    StableDiffusionControlNetImg2ImgPipeline,
    DPMSolverMultistepScheduler,
    EulerDiscreteScheduler
)

# -------- CONFIG -------- #
BASE_MODEL = "SG161222/Realistic_Vision_V5.1_noVAE"
CONTROLNET_MODEL = "monster-labs/control_v1p_sd15_qrcode_monster"

# -------- VAE & CONTROLNET -------- #
vae = AutoencoderKL.from_pretrained(
    "stabilityai/sd-vae-ft-mse",
    torch_dtype=torch.float16
)

controlnet = ControlNetModel.from_pretrained(
    CONTROLNET_MODEL,
    torch_dtype=torch.float16
)

# -------- MAIN PIPELINE -------- #
main_pipe = StableDiffusionControlNetPipeline.from_pretrained(
    BASE_MODEL,
    controlnet=controlnet,
    vae=vae,
    torch_dtype=torch.float16,
    safety_checker=None,   # safety checker disabled
    feature_extractor=None
).to("cuda")

# -------- IMG2IMG PIPELINE -------- #
image_pipe = StableDiffusionControlNetImg2ImgPipeline(**main_pipe.components)

# -------- SCHEDULERS -------- #
SAMPLER_MAP = {
    "DPM++ Karras SDE": lambda config: DPMSolverMultistepScheduler.from_config(
        config, use_karras=True, algorithm_type="sde-dpmsolver++"
    ),
    "Euler": lambda config: EulerDiscreteScheduler.from_config(config),
}

# -------- GRADIO DEMO -------- #
def generate(prompt, control_image, strength=0.8, guidance=7.0, steps=30, sampler="Euler"):
    scheduler = SAMPLER_MAP[sampler](image_pipe.scheduler.config)
    image_pipe.scheduler = scheduler

    image = image_pipe(
        prompt=prompt,
        image=control_image,
        strength=strength,
        guidance_scale=guidance,
        num_inference_steps=steps
    ).images[0]
    return image

with gr.Blocks() as demo:
    gr.Markdown("# ✨ Illusion Diffusion (Fixed)")

    with gr.Row():
        prompt = gr.Textbox(label="Prompt")
        control_image = gr.Image(type="pil", label="Control Image")

    with gr.Row():
        strength = gr.Slider(0.0, 1.0, value=0.8, label="Strength")
        guidance = gr.Slider(1.0, 15.0, value=7.0, label="Guidance Scale")
        steps = gr.Slider(5, 50, value=30, step=1, label="Steps")
        sampler = gr.Dropdown(list(SAMPLER_MAP.keys()), value="Euler", label="Sampler")

    btn = gr.Button("Generate")
    output = gr.Image()

    btn.click(
        fn=generate,
        inputs=[prompt, control_image, strength, guidance, steps, sampler],
        outputs=output
    )

demo.queue().launch()