Awesome

EMA-VFI-WebUI - AI-Based Movie Restoration

frame-interpolation

🎬 Windows 11 example install steps 4/20/2024

💥 See more samples in the Samples Showcase

Example - Interpolated Frames

Example - GIF to MP4 (frame size X4, frame rate X8)	Example - Original GIF
https://user-images.githubusercontent.com/825994/224548062-4cad649c-5cdb-4f66-936d-e2296eb0fbc8.mp4

Example - Resyntheszed Video (YouTube)
https://youtube.com/shorts/lKtY2CHqA98?feature=share
Upper: 8MM footage with heavy dirt and noise
Lower: Same footage after using Resynthesize Video

🎬 EMA-VFI-WebUI Features
➗ Frame Interpolation	Restore Missing Frames, Reveal Hidden Motion
🔎 Frame Search	Synthesize Between Frames At Precise Times
🎈 Video Inflation	Create Super Slow-Motion
💕 Resynthesize Video	Create a Complete Set of Replacement Frames
🪄 Frame Restoration	Restore Adjacent Missing / Damaged Frames
🔬 Video Blender	Project-Based Movie Restoration
📁 File Conversion	Convert between PNG Sequences and Videos
🔢 Resequence Files	Renumber for Import into Video Editing Software
🎞️ Change FPS	Convert any FPS to any other FPS
💎 GIF to MP4	Convert Animated GIF to MP4 in one click
📈 Upscale Frames	Use Real-ESRGAN to Enlarge and Clean Frames

Set Up For Use

Get EMA-VFI working on your local system

See their repo at https://github.com/MCG-NJU/EMA-VFI
I run locally with:
- Anaconda 23.1.0
- Python 3.10.9
- Torch 1.13.1
- Cuda 11.7
- NVIDIA RTX 3090
- Windows 11

Clone this repo in a separate directory and copy all directories/files on top of your working EMA-VFI installation

This code makes no changes to their original code (but borrows some) and causes no conflicts with it
It shouldn't introduce any additional requirements over what EMA-VFI, Gradio-App and Real-ESRGAN need

If it's set up properly, the following command should write a new file images/image1.png using default settings

python interpolate.py

Alternate Set Up / Development

Get EMA-VFI working on your local system

See their repo at https://github.com/MCG-NJU/EMA-VFI
I run locally with:
- Anaconda 23.1.0
- Python 3.10.9
- Torch 1.13.1
- Cuda 11.7
- NVIDIA RTX 3090
- Windows 11

Clone this repo to a directory in which you intend to use the app and/or develop on it
Copy the following directories and files from your working EMA-VFI installation to this directory:

benchmark
ckpt
model
config.py
dataset.py
Trainer.py

If it's set up properly, the following command should write a new file images/image1.png

python interpolate.py

Real-ESRGAN Add-On Set Up

The GIF to MP4 feature uses Real-ESRGAN to clean and upscale frames

Get Real-ESRGAN working on your local system

See their repo at https://github.com/xinntao/Real-ESRGAN

Clone their repo to its own directory and follow their instructions for local setup
Copy the realesrgan directory to your EMA-VFI-WebUI directory

The Real-ESRGAN 4x+ model (65MB) will automatically download on first use

FFmpeg Set Up

A few features rely on FFmpeg being available on the system path

Download FFmpeg

Starting Web UI Application

The application can be started in any of these ways:

webui.bat
python webui.py
- Command line arguments
  - --config_path path path to alternate configuration file, default config.yaml
  - --verbose enables verbose output to the console, default False

Using Web UI

Wiki - Quick Start Guide

All Features

Wiki - Home

Command Line Tools

The core feature have command-line equivalents

Wiki - Command Line Tools

App Configuration

Wiki - App Configuration

Samples Showcase

Acknowledgements

Thanks! to the EMA-VFI folks for their amazing AI frame interpolation tool

https://github.com/MCG-NJU/EMA-VFI

Thans! to the Real-ESRGAN folks for their wonderful frame restoration/upscaling tool

https://github.com/xinntao/Real-ESRGAN

Thanks! to the stable-diffusion-webui folks for their great UI, amazing tool, and for inspiring me to learn Gradio

https://github.com/AUTOMATIC1111/stable-diffusion-webui

Thanks to Gradio for their easy-to-use Web UI building tool and great docs