MiniPx
Add to ChromePricingBlogAccount
No. 05 — AI Tools · Cloud HDPro · on our GPU

Restore, colorize, one old photo.

One upload, two models, one job: the faces are rebuilt first and colour is put on the restored result. It is the convenient route rather than the cheap one, and this page sets out exactly what it costs and when to split it in two instead.

Pro · Cloud HDDeleted on completionUp to 25 MB · 24 MP
Restore and add colour in one runCloud HD · our GPU

Pro to restore your photo →

How it works

  1. Pick the monochrome photograph — JPEG, PNG or WebP, up to 25 MB and 24 megapixels.
  2. Our GPU restores the faces and checks each one against your original. If the result drifts too far from the person in your photo, the run is refused rather than handing back an invented face — and you are not charged.
  3. Switch the colour step on before running to get a colourised PNG back at the same resolution; that makes the job one HD run rather than an ordinary one.

This page is the photo restorer, tuned to its chained colour option and the arithmetic behind it.

All Image Tools

AI Tools

Pro Tools

Compress

Convert

Resize

Compress to Size

Compress for Platform

Tools & Utilities

By Device

Compare

Solutions by use case

For Developers & Web

For Sellers & Ecommerce

For Job Applicants & Govt Forms

For Creators & Social

For Work & Documents

Restore and Colorize an Old Photo — Pro, on our server

The chained run is one upload and two models. Your photograph goes to our GPU once. RestoreFormer++ finds each face, rebuilds it at high resolution and blends it back; then DDColor takes that restored image and predicts colour for it. What comes back is a single colourised PNG at exactly the resolution you uploaded. There is no second upload, no second wait for a queue and no intermediate file to manage — which is the entire argument for doing it this way. The identity check still applies to the restoration half, so a job whose rebuilt face has drifted from the person in your photograph is refused rather than returned, and a refused job is not charged.

The order inside that job is not arbitrary. A colorizing model works by recognising what things are — this region is skin, that is foliage, this is a wooden door — and it recognises them from structure. Hand it a face that has been rebuilt and it has clean edges to work against; hand it a soft grey oval and the colour tends to bleed across boundaries the model could not find. Restoring first also means the colour lands on the version you actually intend to keep. It is the same reason the faded prints page puts restore before colorize and enlargement last, and why the main restorer describes the sequence the same way.

Now the arithmetic, because convenience here has a real price. Cloud HD gives Pro Monthly 150 runs a month and Pro Annual 200, resetting on the 1st in UTC with no rollover and a ceiling of 30 an hour. Inside that sits a smaller pot: up to 10 HD runs a month, shared between object removal, Cloud HD upscaling and exactly this chained restore-plus-colour job. So the chained version spends one of ten, while restoring here as an ordinary run and then putting the downloaded PNG through the standalone colorizer spends two of a hundred and fifty and none of the ten. If you are working through a box of prints, the two-step route is almost always the right one; if you are doing a single photograph and would rather not shuttle files about, chain it. Both produce the same kind of result. For photographs that are already sharp and only need colour, skip restoration entirely and go straight to colorizing old photos.

What the colour is, and where the file goes. The colour is plausible, not verified. DDColor produces what is statistically reasonable for the scene in front of it, and it has no way of knowing what colour that coat, that car or that regimental uniform actually was. Skin, sky, grass and timber are usually convincing; dyed fabric is a well-informed guess. Keep the monochrome scan — see restoring black and white photos for why the original stays the record. As for privacy: this is Cloud HD, the Pro-only, opt-in feature set that runs on our server because these models are far too heavy for a browser. Your photograph is encrypted in transit, uploaded to our server for that run and deleted when the run finishes, and in all cases within 24 hours; it is processed in the United States on Cloudflare R2 storage and a Modal GPU, and never used to train anything. The model that ran is named on the job and can change — you can always see which one processed yours. Cropping, converting and compressing stay on your own device.

How it works

  1. Check the photograph is genuinely monochrome: A faded colour print needs a white-balance correction on your own device instead, and a negative needs inverting before anything else.
  2. Upload it once: JPEG, PNG or WebP, up to 25 MB and 24 megapixels. Crop to the print edge first so the faces occupy more of the frame.
  3. Turn on the colour step before you run it: The job then rebuilds the faces and adds colour to the restored result in a single pass, and is billed as one HD run.
  4. Keep the monochrome original: Save the colourised PNG as a derivative beside your untouched scan. The colour is an interpretation; the scan is the record.

Frequently asked questions

Why does chaining the colour step make it an HD run?
Because the job then does two model passes on your photograph instead of one: RestoreFormer++ rebuilds the faces, and DDColor then puts colour on the result. HD runs are the smaller pot inside your allowance — up to 10 a month out of 150 on Pro Monthly or 200 on Pro Annual — shared with object removal and Cloud HD upscaling.
Can I do the two steps as separate runs instead?
Yes, and it is often the smarter trade. Restore here as an ordinary run, download the result, then put that PNG through the standalone colorizer as a second ordinary run. Two ordinary runs spend nothing from the pot of 10 HD runs, which you may want to keep for object removal. The chained version buys convenience, not quality.
Why restore first rather than colour first?
Because a colour model reads structure to decide what things are, and rebuilt faces give it far more structure to read than a soft grey oval does. Colour laid over blur tends to smear across the boundaries it cannot see. Restoring first also means the colour is applied to the version you intend to keep, rather than to an intermediate you are about to replace.
Is the colour it produces accurate?
It is plausible, not verified. DDColor predicts colours that are statistically reasonable for the scene it sees; it has no access to what your grandmother’s coat was actually made of. Skin, sky, foliage and wood are usually convincing. Dyed fabric, painted vehicles and uniforms are educated guesses. Treat the result as an interpretation and keep the monochrome original.
Will it colorize a photo that is already in colour, or a negative?
A faded colour print is not what this step is for — that needs a levels or white-balance correction in your own editor, which costs nothing and no upload. Negatives are firmly out of scope: nothing here inverts a negative, so scan and invert it yourself first, then restore the resulting positive image.
Does a chained colorize run mean two uploads?
No. One upload, one job, one result. Your photograph is uploaded to our server for that run and deleted when the run finishes, and in all cases within 24 hours, whether the job did one model pass or two. It is encrypted in transit, processed in the United States on Cloudflare R2 storage and a Modal GPU, and never used to train anything.