handler.py

import base64
import io
from PIL import Image
import torch
from diffusers import StableDiffusionXLPipeline
from typing import Any, Dict


class EndpointHandler:
    def __init__(self, model_dir: str, **kwargs):
        print("🔥 Initializing Juggernaut XL Handler (Prompt + Optional Image)...")

        # Load XL model from your big repo
        self.pipe = StableDiffusionXLPipeline.from_pretrained(
            "Gjm1234/juggernaut-sfw",
            torch_dtype=torch.float16,
            use_safetensors=True
        ).to("cuda")

        self.pipe.enable_attention_slicing()

        print("✅ Pipeline loaded successfully.")

    def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]:

        # Must receive `inputs`
        if "inputs" not in data:
            return {"error": "Body must contain 'inputs' object"}

        inputs = data["inputs"]

        prompt = inputs.get("prompt", None)
        if not prompt:
            return {"error": "prompt is required"}

        num_images = inputs.get("num_images", 10)
        image_b64 = inputs.get("image", None)

        init_image = None
        if image_b64:
            try:
                img_bytes = base64.b64decode(image_b64)
                init_image = Image.open(io.BytesIO(img_bytes)).convert("RGB")
            except Exception as e:
                return {"error": f"Invalid image data: {str(e)}"}

        # Run txt2img OR img2img depending on whether image was sent
        if init_image is None:
            print("🎨 Running TEXT → IMAGE")
            output = self.pipe(
                prompt=prompt,
                num_images_per_prompt=num_images
            )
        else:
            print("🎨 Running IMAGE → IMAGE")
            output = self.pipe(
                prompt=prompt,
                image=init_image,
                strength=0.6,
                num_images_per_prompt=num_images
            )

        images = output.images

        # convert to base64 array
        results = []
        for img in images:
            buffered = io.BytesIO()
            img.save(buffered, format="PNG")
            img_b64 = base64.b64encode(buffered.getvalue()).decode("utf-8")
            results.append(img_b64)

        print(f"✅ Returning {len(results)} images.")
        return {"images": results}