No. 06 — AI Tools · Cloud HDPro · on our GPU

Historical photos, in plausible color.

Colour makes a historical photograph land differently — closer, more like a memory than a document. It also makes a claim the negative never made. This page is about getting the first without pretending the second is free. Uploads run once on our Cloud HD server, are deleted on completion, and are never used to train anything.

Pro · Cloud HDDeleted on completionUp to 25 MB · 24 MP
Black-and-white photograph of a long-haired tabby cat, before colorization
Sample — processed with ddcolor-modelscope 8565d1f7c686
Colorize a historical photoCloud HD · our GPU

How it works

  1. Upload the archive scan — JPEG, PNG or WebP, up to 25 MB and 24 megapixels.
  2. DDColor infers colour on our GPU for that single run; the input is then deleted.
  3. Download a same-size PNG and caption it as digitally colorized. One Cloud HD run.

This page is the photo colorizer, tuned to documentary and archive work where the caption matters as much as the image.

All Image Tools

AI Tools

Solutions by use case

Colorize Historical PhotosPro, on our server

Colorization is an interpretation of a document, not a correction of it. The monochrome photograph is the primary source; the coloured version is a reading of it made in 2026 by a neural network with no knowledge of your subject. DDColor was trained on modern colour photographs and their greyscale conversions, so what it produces is the most statistically likely colour for each region, given how those greys usually arise. That is genuinely useful — it is how a skilled restorer reasons too — but it is a probability, not a record. A researcher who keeps that distinction explicit gets all of the communicative power and none of the trouble that follows a confident, invented, uncited colour.

Write the caption before you publish the picture. The convention that works is short and blunt: name the original and its holding collection, then say plainly that colour was added digitally, in what year, and with which model. DDColor is the model here, and the tool always shows you which model ran; the model can change, so recording the one that produced a particular file is worth the ten seconds it takes. If a specific colour in the frame is actually documented — a regimental facing colour, a livery, a flag — correct it by hand afterwards and say in the caption that you did. Where the subject is a uniform, the military photographs page goes into that in more detail.

The tool is deliberately conservative about detail. It changes chroma and nothing else, returning a PNG at exactly the resolution you uploaded — up to 25 MB and 24 megapixels. Nothing is sharpened, nothing is filled in, no scratch is painted out and no missing corner is reconstructed. Damage that was in the scan is in the output, in colour. If a print genuinely needs its faces recovered before colour will read at all, the face restorer is the separate step for that and should be disclosed separately; if the master scan is too small to reproduce, enlarge afterwards with the on-device upscaler rather than asking the colorizer for resolution it cannot give. The wider argument for keeping these steps distinct is on the old photos page.

Where archive material goes when you press the button. Colorization is a Cloud HD tool, which means it is the exception to the rule that MiniPx never uploads your images: the file goes over an encrypted connection to our server, is processed in the United States on Cloudflare R2 storage and a Modal GPU, and is deleted when the run finishes and in all cases within 24 hours. It is never used to train anything and it is never shown to anyone. If you are working under an institution’s licence terms, that retention window is the fact they will want in writing — and everything else you do on MiniPx, from cropping to converting, happens without the file leaving your machine at all.

How it works

  1. Record the provenance first: Note collection, reference number and date against the monochrome scan before you touch it. The colour version inherits nothing automatically.
  2. Work from the best scan available: Use the highest-quality master you are permitted to use, up to 25 MB and 24 megapixels, rather than a compressed web thumbnail.
  3. Run DDColor once: Upload the file; the model runs on our Cloud HD GPU and returns a PNG at the same resolution as your input.
  4. Caption the result as colorized: Attach a one-line note naming the model and the year, and keep the monochrome master as the citable source.

Frequently asked questions

Is a colorized historical photo evidence of anything?
Only of what the photograph already showed. The colour layer is inference, not testimony: DDColor predicts the colours that most plausibly produce those grey values, drawing on modern photographs it was trained on. Treat the output as an illustration, cite the monochrome original as the source, and never let a colorized frame stand alone in a footnote.
How should I caption a colorized archive image?
Name the original, then the intervention. Something like “Original photograph, [collection, date]; colour added digitally in 2026 using an AI model (DDColor). Colours are inferred and not historically verified.” One sentence removes every ambiguity a reader might otherwise resolve in your favour by accident.
Can I publish the colorized version of an archive photograph?
That is a rights question, not a tool question. Colorizing does not create a new copyright position for you and does not clear the underlying image, so check the holding institution’s terms exactly as you would before reproducing the monochrome scan. What we can tell you is that MiniPx claims nothing over your output and stores nothing after the run.
Does the model know about period-specific colours?
No. It has no timeline, no captions and no knowledge of your subject. It has learned that foliage tends to be green and skin tends to be warm, and it applies those expectations uniformly, so a 1910 street and a 2010 street get the same reasoning. Period-specific detail — livery, uniforms, signage — is where you should expect it to guess.
Will it clean up an archive scan while it colorizes?
No. Colorization only changes chroma; the light-and-dark structure passes through untouched, so dust, scanner banding, tape marks and emulsion damage all reappear in colour. That is deliberate — the tool does not invent detail that the original does not contain, which is exactly the property archival work needs.