Skip to content
Get packs

DeOldify

Uses deep learning to add color to old black-and-white photos and film, and it is now archived.

GitHub Download source
Stars
18.5k
Forks
2.6k
Licence
MIT
Language
Python
Last update
2.0 years ago
Type
AI model / agent

This repository is archived and no longer maintained.

What it is

DeOldify is an open-source project that uses deep learning, a type of AI trained on many images, to colorize black-and-white photos and footage and to restore old images. It handles still images and video, where each frame is colorized separately. The author has archived the repository, so it is no longer maintained, though the code and models remain available.

  • Colorize a scanned black-and-white photo using the linked DeepAI or Colab notebooks
  • Colorize short silent film or old video footage, using the video Colab notebook
  • Run the stable image model if the default artistic model produces glitches
  • Try colorization in a browser through the linked ONNX-based DeOldify-on-Browser project

Great for

  • People who want to colorize old family or archival photos without paying for a service
  • Hobbyists comfortable running Python-based machine learning code
  • Researchers or students interested in how GAN-based image colorization works
  • Users who want a browser or desktop option described in the README, such as the linked community projects

Not ideal for

  • Anyone needing ongoing updates or bug fixes, because the repository is archived and the author says he no longer has time to maintain it
  • People expecting historically accurate colors, since the README admits accuracy is a major challenge
  • Users who need the most advanced version, since the README points to MyHeritage In Color as the most advanced, a paid hosted service outside this repo

How to get it

  1. 1

    Clone and build from source

    You'll need: Python · Git

    Clone the repository

    git clone https://github.com/jantic/DeOldify.git
  2. 2

    Run it

    • Open one of the linked Colab notebooks, such as ImageColorizerColab.ipynb, in your browser
    • Upload a black-and-white image or video
    • Run the notebook cells in order to colorize it
    • Download the result
  3. 3

    Get the most out of it

    • Try the stable image model if the artistic model shows glitches, as the README suggests
    • Use a higher render_factor for more consistent colors, according to the README's stability notes
    • Use resnet101 instead of resnet34 as the generator backbone for more consistent skin and object rendering
    • Watch the linked tutorial video for Colab usage instructions

Downloads

Source code (.zip)master branch · straight from GitHub

Strengths

  • Widely used, with a large community and many linked tools built on it
  • Several no-install options are linked, including Colab notebooks and a browser version
  • The README says the video results are stabilized by using NoGAN training and higher render resolution

Trade-offs

  • Archived and unmaintained, so issues may not be fixed and the models are dated, as the author himself notes
  • Colors are not guaranteed to be historically accurate, as the README's Golden Gate Bridge example shows
  • Running it locally takes setup work and a suitable computer, and the README's local instructions were truncated so the exact requirements are unclear

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.