handler.py

import base64
import io
import cv2
import numpy as np
from PIL import Image
import torch
from diffusers import AutoPipelineForInpainting

class EndpointHandler:
    def __init__(self, path=""):
        print("[INIT] Loading Nano Banana SDXL Inpainting pipeline...")

        # Load Nano Banana (SDXL fine-tuned)
        self.pipe = AutoPipelineForInpainting.from_pretrained(
            "SG161222/RealVisXL_V4.0_Nano-Banana",
            torch_dtype=torch.float16,
            variant="fp16"
        ).to("cuda")

        # Default high-level removal instruction
        self.default_prompt = "remove text captions, natural background, realistic restoration"
        print("[READY] Nano Banana model loaded successfully.")

    def _decode_image(self, b64_image):
        img_bytes = base64.b64decode(b64_image)
        img = Image.open(io.BytesIO(img_bytes)).convert("RGB")
        return img

    def _encode_image(self, pil_img):
        buf = io.BytesIO()
        pil_img.save(buf, format="PNG")
        return base64.b64encode(buf.getvalue()).decode("utf-8")

    def __call__(self, data):
        if "image" not in data["inputs"]:
            raise ValueError("Missing 'image' field in inputs")

        prompt = data["inputs"].get("prompt", self.default_prompt)

        # Decode base64 → PIL
        img_pil = self._decode_image(data["inputs"]["image"])

        print(f"[PROCESS] Running Nano Banana with prompt: '{prompt}'")

        # Inpaint the whole image (no mask — full generative clean-up)
        result = self.pipe(
            prompt=prompt,
            image=img_pil,
            mask_image=None,
            guidance_scale=3.0,
            strength=0.85,
            num_inference_steps=25
        ).images[0]

        # Encode result back to base64
        cleaned_b64 = self._encode_image(result)

        return {"image": cleaned_b64}