Restoration face: bring old photos back

There is a photo of your great-grandfather in the family album: the face has blurred, the corners are worn and everything is black and white. It feels awkward to show it to relatives, and a studio asks a lot for retouching. Restoration face handles that job inside Telegram. 📷

The bot does three things: it restores faces, increases image resolution and colourises black-and-white photos. All of it runs on neural network models that the developer connected to the chat.

Three tools in one chat

The bot's description lists three directions. They fix different problems, so pick one to match the photo in front of you.

• Face restoration. The models redraw sharp features where the original is a smudge. It works best on portraits.
• Upscaling. A small picture becomes a larger one, with crisper edges and textures. Handy if you plan to print the photo or use it as an avatar.
• Colourisation. A black-and-white or sepia frame gets colour. The model picks the shades itself, so the colour of a coat or the sky may differ from the real one.

Getting a better result

A neural network works with what it sees, and a cleaner source gives a more believable output. A few habits worth trying.

1. Send the biggest file. Use the original or the best scan, not a forwarded copy that a messenger already compressed.
2. Do not crop too hard. For faces, keeping the whole frame gives the model more context.
3. Run the modes one by one. Restore the face first, upscale next and colourise at the end.
4. Compare with the original. The model sometimes invents details, so check important photos against the source.

One thing to know before you start: each model has its own character. One handles portraits better, another does scenery and buildings. If the first result disappoints, try again with another mode or another source file.

Queue, help and supporting the author

The bot has three commands. The /status command shows the queue: processing runs on shared hardware, so at busy hours you may wait.

The /help command gives a short guide to the modes. The /donate command is for voluntary support of the author. Paying is not required to understand how the bot works, but it is a way to say thanks if the service saved a photo for you.

Limits worth knowing

Neural face restoration does not bring back what was lost. It guesses how the face might have looked. That is why a person in the photo can end up a bit smoother and younger than they were.

Colourisation does not know the real colours either. For a family archive that is fine. For historical work it is not reliable.

Finally, when you upload a photo you send it to someone else's server. Do not send documents, pictures of children without a parent's consent, or anything you would not want in other hands. 🔒

Albums, prints and tribute pages

People sorting family archives, preparing photos for print, making a tribute page, or just wanting to see a grandfather in colour. It will not replace professional hand restoration, but for a quick result it does the job.