app.py

# app.py
import gradio as gr
import torch, io, base64
from PIL import Image
from diffusers import StableDiffusionImg2ImgPipeline
from vtoonify_model import load_vtoonify  # see below

# Load models
pipe_ghibli = StableDiffusionImg2ImgPipeline.from_pretrained(
    "nitrosocke/Ghibli-Diffusion", torch_dtype=torch.float16
).to("cuda")  # Ghibli-style fine-tuned SD :contentReference[oaicite:1]{index=1}

pipe_vtoonify = load_vtoonify().to("cuda")  # cartoonization model loader

# Helpers for base64 conversion
def pil_to_b64(img: Image.Image) -> str:
    buf = io.BytesIO()
    img.save(buf, format="PNG")
    return base64.b64encode(buf.getvalue()).decode()

def b64_to_pil(b64: str) -> Image.Image:
    data = base64.b64decode(b64)
    return Image.open(io.BytesIO(data)).convert("RGB")

# Core processor
def run_effect(input_b64: str, effect: str) -> dict:
    img = b64_to_pil(input_b64)
    if effect == "ghibli":
        out = pipe_ghibli(prompt="ghibli style", image=img, strength=0.5, guidance_scale=7.5).images[0]
    else:
        out = pipe_vtoonify(img)
    return {"output_b64": pil_to_b64(out)}

@gr.utils.decorators.thread_safe()
@spaces.GPU  # enables GPU on ZeroGPU Infra
def api_process(input_b64, effect):
    return run_effect(input_b64, effect)

def gradio_process(img: Image.Image, effect: str) -> Image.Image:
    # Reuse logic, bypass base64
    in_b64 = pil_to_b64(img)
    return b64_to_pil(run_effect(in_b64, effect)["output_b64"])

with gr.Blocks() as demo:
    gr.Markdown("# Ghibli & VToonify Effects 🎨")

    with gr.Tab("Web UI"):
        inp = gr.Image(type="pil", label="Upload Image")
        eff = gr.Radio(["ghibli", "vtoonify"], label="Effect")
        btn = gr.Button("Apply Effect")
        out = gr.Image(label="Result")
        btn.click(gradio_process, [inp, eff], out)

    with gr.Tab("API (base64)"):
        inp_b64 = gr.Textbox(lines=4, label="Input Image (base64)")
        eff2 = gr.Radio(["ghibli", "vtoonify"], label="Effect")
        btn2 = gr.Button("Run API")
        out_b64 = gr.Textbox(lines=4, label="Output Image (base64)")
        btn2.click(api_process, [inp_b64, eff2], out_b64)

demo.launch()