import os
os.system("git clone https://github.com/google-research/frame-interpolation")
import sys
sys.path.append("frame-interpolation")
import numpy as np
import tensorflow as tf
import mediapy
from PIL import Image
from eval import interpolator, util
import tensorflow as tf
import gradio as gr


_UINT8_MAX_F = float(np.iinfo(np.uint8).max)

from huggingface_hub import snapshot_download
model = snapshot_download(repo_id="akhaliq/frame-interpolation-film-style")

interpolator = interpolator.Interpolator(model, None)

batch_dt = np.full(shape=(1,), fill_value=0.5, dtype=np.float32)

def predict(frame1, frame2, times_to_interpolate):
    img1 = frame1
    img2 = frame2
    if not img1.size == img2.size:
        img1 = img1.crop((0, 0, min(img1.size[0], img2.size[0]), min(img1.size[1], img2.size[1])))
        img2 = img2.crop((0, 0, min(img1.size[0], img2.size[0]), min(img1.size[1], img2.size[1])))
        frame1 = 'new_frame1.png'
        frame2 = 'new_frame2.png'
        img1.save(frame1)
        img2.save(frame2)

    input_frames = [str(frame1), str(frame2)]

    frames = list(
        util.interpolate_recursively_from_files(
            input_frames, times_to_interpolate, interpolator))
    ffmpeg_path = util.get_ffmpeg_path()
    mediapy.set_ffmpeg(ffmpeg_path)
    out_path =  "out.mp4"
    mediapy.write_video(str(out_path), frames, fps=30)
    return out_path

gr.Interface(predict,[gr.inputs.Image(type='pil'),gr.inputs.Image(type='pil'),gr.inputs.Slider(minimum=2,maximum=5,step=1)],"playable_video").launch(enable_queue=True)