from typing import Dict, List, Any
from PIL import Image
from io import BytesIO
import torch
import base64
from diffusers import StableDiffusionInstructPix2PixPipeline, EulerAncestralDiscreteScheduler

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

class EndpointHandler():
    def __init__(self, path=""):
        model_id = "timbrooks/instruct-pix2pix"
        self.pipe = StableDiffusionInstructPix2PixPipeline.from_pretrained(model_id, torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32, safety_checker=None)
        self.pipe.to(device)
        self.pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(self.pipe.scheduler.config)

    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
        """
       data args:
            inputs (:obj:`string`)
            parameters (:obj:)
      Return:
            A :obj:`string`:. image string
        """


        image_data = data.pop('inputs', data)
        # decode base64 image to PIL
        image = Image.open(BytesIO(base64.b64decode(image_data)))

        parameters = data.pop('parameters', [])
        prompt = parameters.pop('prompt', None)
        negative_prompt = parameters.pop('negative_prompt', None)
        num_inference_steps = parameters.pop('num_inference_steps', 10)
        image_guidance_scale = parameters.pop('image_guidance_scale', 1.5)
        guidance_scale = parameters.pop('guidance_scale', 7.5)


        images = self.pipe(
            prompt,
            image = image,
            negative_prompt = negative_prompt,
            num_inference_steps = num_inference_steps,
            image_guidance_scale = image_guidance_scale,
            guidance_scale = guidance_scale
        ).images

        return images[0]