app.py

import warnings
import gradio as gr
from transformers import pipeline
import io, base64
from PIL import Image
import numpy as np
import tensorflow as tf
import mediapy
import os
import sys
from huggingface_hub import snapshot_download

#CREDIT: this demo is based *heavily* on https://huggingface.co/spaces/osanseviero/latent-video

with warnings.catch_warnings():
  warnings.simplefilter('ignore')
  image_gen = gr.Interface.load("spaces/multimodalart/latentdiffusion")

  os.system("git clone https://github.com/google-research/frame-interpolation")
  sys.path.append("frame-interpolation")
  from eval import interpolator, util
  
ffmpeg_path = util.get_ffmpeg_path()
mediapy.set_ffmpeg(ffmpeg_path)

model = snapshot_download(repo_id="akhaliq/frame-interpolation-film-style")
interpolator = interpolator.Interpolator(model, None)


def generate_images(text, width=256, height=256, steps=50, num_images=2,
                    diversity=4):

    image_bytes = image_gen(text, steps, width, height, num_images, diversity)
    
    # Algo from spaces/Gradio-Blocks/latent_gpt2_story/blob/main/app.py
    generated_images = []
    for image in image_bytes[1]:
        image_str = image[0]
        image_str = image_str.replace("data:image/png;base64,","")
        decoded_bytes = base64.decodebytes(bytes(image_str, "utf-8"))
        img = Image.open(io.BytesIO(decoded_bytes))
        generated_images.append(img)
        
    return generated_images


def generate_interpolation(text, fps=7, steps=4):    
  images = []
  frames = []
  for i, t in enumerate(text.split(', ')):
    print(f'image {i}: {t.lower().strip()}', end='...')
    images.extend(generate_images(t.lower().strip()))
    print('done!')
  
    frames.append(f'frame_{i}.png')
    images[-1].save(frames[-1])

  vid = list(util.interpolate_recursively_from_files(frames, steps, interpolator))
  mediapy.write_video("out.mp4", vid, fps=fps)
  return "out.mp4"

demo = gr.Blocks()

with demo:
  text = gr.Textbox(placeholder='human, human brain, brain in a computer, humanoid robot', label='Input a comma-separated list of terms:')
  button = gr.Button("Generate Video")
  output = gr.Video(label="Generated Video")
        
  button.click(fn=generate_interpolation, inputs=text, outputs=output)  

demo.launch(debug=True, enable_queue=True)