Free
$0
- 3 photos per day
- Low resolution preview
- Small site mark
Checkout opens when billing is connected (local build).
Colorization is a prediction problem, not a paint-by-numbers conversion. Knowing which parts are predicted keeps the result useful.
Upload a black and white photo. No account needed for your first 3 photos each day.
One quick human check before each free photo — it stops scripts from burning through the pool.
Demo mode — output is a placeholder
First, segmentation: which pixels belong to skin, hair, fabric, sky, water or foliage. Second, palette: a statistical guess conditioned on era, lighting and subject. Third, blending: keeping edges and specular highlights from turning into color fringing.
Those are the three places where a result looks wrong. A blue sky on an indoor studio portrait, a greenish cast on skin, or a hard color line across a shirt all come from one of those decisions, and re-running with a different crop can change them.
Send the largest version of the scan you have. A 700-pixel-wide scan leaves almost nothing for the face restoration step, so eyes stay soft and the color boundaries get muddy.
Stay in JPG, PNG or WEBP and avoid heavy JPEG artefacts from repeated re-saving: the model reads blocky noise as texture and colors it.
JPG, PNG or WEBP. Faces, wedding portraits and family archive scans give the best results.
A face restoration model rebuilds eyes and mouths, a super-resolution model doubles the detail, and a colorization model paints natural skin and fabric tones.
Free previews are low resolution with a site mark. HD credits remove both limits.
$0
Checkout opens when billing is connected (local build).
$10.00
$10.00 per photo
$25.00
$5.00 per photo
$30.00
$3.00 per photo
$39.00
$3.25 per photo
This build processes one photo per job. A queue view for batches is planned with the subscription tier.
Yes. Sepia scans still have a single luminance channel plus a warm cast; the model normalises the cast first, then predicts color, which usually works better than starting from pure greyscale.