handler.py

from typing import Dict
import torch
from diffusers import DiffusionPipeline
from compel import Compel
from io import BytesIO
import base64

class EndpointHandler:
    def __init__(self, path: str = ""):
        # Load base FLUX pipeline
        self.pipe = DiffusionPipeline.from_pretrained(
            "black-forest-labs/FLUX.1-dev",
            torch_dtype=torch.float16,
            variant="fp16",
        )

        # Load your LoRA weights from the repo
        self.pipe.load_lora_weights("./c1t3_v1.safetensors")

        # Move to GPU if available
        if torch.cuda.is_available():
            self.pipe.to("cuda")
        else:
            self.pipe.to("cpu")

        # Optional: enable memory optimization
        self.pipe.enable_model_cpu_offload()

        # Initialize Compel (prompt parser for FLUX)
        self.compel = Compel(tokenizer=self.pipe.tokenizer, text_encoder=self.pipe.text_encoder)

    def __call__(self, data: Dict[str, str]) -> Dict:
        # Get prompt from request data
        prompt = data.get("prompt", "")
        if not prompt:
            return {"error": "No prompt provided."}

        # Generate prompt conditioning using Compel
        conditioning = self.compel(prompt)

        # Generate image using FLUX + LoRA
        image = self.pipe(prompt_embeds=conditioning).images[0]

        # Convert image to base64 string for API response
        buffer = BytesIO()
        image.save(buffer, format="PNG")
        base64_image = base64.b64encode(buffer.getvalue()).decode("utf-8")

        return {"image": base64_image}