app.py

import gradio as gr
import torch
from diffusers import StableDiffusionImg2ImgPipeline
from torchvision import transforms
from PIL import Image
import io
import base64
import spaces
from functools import lru_cache

# Base64 utilities
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:
    return Image.open(io.BytesIO(base64.b64decode(b64))).convert("RGB")

# ✅ Cached Model Loaders (ZeroGPU Safe)
@lru_cache(maxsize=2)
def load_ghibli_model():
    pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
        "nitrosocke/Ghibli-Diffusion",
        torch_dtype=torch.float16,
        use_safetensors=True
    ).to("cuda")
    return pipe

@lru_cache(maxsize=2)
def load_animegan_model():
    model = torch.hub.load(
        "bryandlee/animegan2-pytorch:main",
        "generator",
        pretrained="face_paint_512_v2"
    ).to("cuda").eval()
    return model

# ✅ Image Processing (Gradio Image Upload)
@spaces.GPU
def process_image(img: Image.Image, effect: str) -> Image.Image:
    if effect == "ghibli":
        pipe = load_ghibli_model()
        out_img = pipe(prompt="ghibli style", image=img, strength=0.5, guidance_scale=7.5).images[0]
    else:
        animegan = load_animegan_model()
        transform = transforms.Compose([
            transforms.Resize((512, 512)),
            transforms.ToTensor()
        ])
        img_tensor = transform(img).unsqueeze(0).to("cuda")
        with torch.no_grad():
            out = animegan(img_tensor)[0].clamp(0, 1).cpu()
        out_img = transforms.ToPILImage()(out)
    return out_img

# ✅ Base64 API Processing
@spaces.GPU
def process_base64(b64: str, effect: str) -> str:
    img = b64_to_pil(b64)
    out_img = process_image(img, effect)
    return pil_to_b64(out_img)

# ✅ Gradio UI
with gr.Blocks() as demo:
    gr.Markdown("# 🎨 Ghibli & AnimeGAN Effects (ZeroGPU Compatible)")

    # Image Upload Tab
    with gr.Tab("Web UI"):
        img_input = gr.Image(type="pil", label="Upload Image")
        effect_choice = gr.Radio(["ghibli", "anime"], label="Select Effect")
        process_btn = gr.Button("Apply Effect")
        img_output = gr.Image(label="Processed Image")
        process_btn.click(process_image, [img_input, effect_choice], img_output)

    # Base64 API Tab
    with gr.Tab("Base64 API"):
        b64_input = gr.Textbox(label="Input Image (Base64)", lines=5)
        effect_choice_b64 = gr.Radio(["ghibli", "anime"], label="Select Effect")
        process_btn_b64 = gr.Button("Run API")
        b64_output = gr.Textbox(label="Output Image (Base64)", lines=5)
        process_btn_b64.click(process_base64, [b64_input, effect_choice_b64], b64_output)

demo.launch()