import gradio as gr
import os

import torch
from diffusers import AutoencoderTiny
from torchvision.transforms.functional import to_pil_image, center_crop, resize, to_tensor

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

model_id = "madebyollin/taesd"
vae = AutoencoderTiny.from_pretrained(model_id, safetensors=True).to(device)


@torch.no_grad()
def encode(image):
    DIM = 512
    processed = center_crop(resize(image, DIM), DIM)
    tensor = to_tensor(processed).unsqueeze(0).to(device)
    latents = vae.encoder(tensor)
    scaled = vae.scale_latents(latents).mul_(255).round_().byte()
    return to_pil_image(scaled[0])


astronaut = os.path.join(os.path.dirname(__file__), "images/6.png")


def app():
    return gr.Interface(encode,
                        gr.Image(type="pil",
                                 mirror_webcam=False,
                                 label='512x512',
                                 value=astronaut),
                        gr.Image(type="pil",
                                 image_mode="RGBA",
                                 label='64x64',
                                 height=256,
                                 width=256
                                 ),
                        examples=[
                            astronaut,
                            os.path.join(os.path.dirname(__file__), "images/7.png"),
                            os.path.join(os.path.dirname(__file__), "images/34.png")
                        ], allow_flagging='never', title='Image Encoder')


if __name__ == "__main__":
    app().launch()