Colorize, don't redraw

October 2026 · 5 min read

There are now two very different kinds of AI that will "colorize" your old photo. One adds color to the photo you gave it. The other draws a new photo that looks like yours. For a snapshot it hardly matters. For your great-grandmother's only portrait, it does.

Two ways to get a color photo

A colorizer separates your photo into lightness (every bit of light and shadow, which carries all the detail: faces, wrinkles, text, film grain) and color. A black-and-white photo has the first and none of the second. The colorizer predicts only the color and lays it over the untouched original lightness. Whatever it gets wrong, it can only get a shade wrong.

A generative image model, like the ones built into chatbots, works from your photo and your prompt and generates a brand-new image. The best models are remarkably faithful, and getting better. But every pixel is newly drawn, so details can drift: a slightly different nose, an extra button, a softened scar, a sign with different letters. OpenAI's own help center notes that edits "may extend beyond the area you selected".

Why it matters for family history

• Faces are what we look at. We recognize people by tiny details, and a near-copy of a face can feel subtly wrong to the people who knew them.
• Photos are evidence. Genealogists read uniforms, shop signs, dates and jewelry in old photos. A redrawn photo can't be used as a source.
• Copies travel. Once a redrawn "restoration" is shared, it can quietly replace the original in a family's memory. Good tools label AI images (OpenAI adds Content Credentials, for example), but labels are easily lost in screenshots.

How to tell which one you're using

1. Compare at full size. Put the original and the result side by side and zoom into an eye, a hand or some text. A colorizer's result has identical detail; a generated one has subtle differences.
2. Check the resolution. A colorizer gives you back the exact size you uploaded. A generated image comes back at the model's own output size, which usually isn't your scan's.
3. Read how it describes itself. "Adds color", "keeps the original" or "never redraws" point to a colorizer. "Generate", "reimagine" or "enhance faces" point to generation.

Where generation is the right tool

Generative tools can do things a colorizer can't: rebuild a missing corner, remove a tear across a face, or turn a damaged print into a clean picture. That's valuable, as long as everyone understands the result is a reconstruction. If you use one, keep the original, label the new version, and don't let it replace the real thing.

What Ungray does

Ungray is a colorizer. It keeps the lightness of every pixel of your photo, at full resolution, and adds only color. If it guesses a color wrong, you can fix it with Recolor: tap the area and pick the right color, and nothing else changes. Upscaling is a separate, optional step for small prints, and it's clearly marked.