How it works

From photograph to finished painting

The analysis runs once, in seconds. Everything after that is instant.

  1. 01

    Drop in your reference

    Any photograph — a portrait, a still life, a devotional subject. 4K and larger are fine, and the original is never altered.

  2. 02

    Choose how many shades

    Start with five broad shades in the free studio. Higher shade counts are planned for paid tiers. Analysis time depends on your image, settings and device.

  3. 03

    Paint one shade at a time

    Mix the recipe, paint where the map says, move to the next. Light to dark, then bring the reference back up and blend the edges by eye.

A session, start to finish

  1. Transfer the drawing. Outline mode gives you the boundary shapes to square up onto canvas — the same shapes you are about to fill.
  2. Mix shade 01. The panel gives you HEX, RGB and Lab, plus the closest mixture from the tubes you own, with the predicted result and how far off it is.
  3. Paint where it says. Only that shade’s areas are visible. Dot mode marks the area without covering your reference, which suits thin passages and glazes.
  4. Work light to dark along the shade strip. Each shade tells you how much of the picture it covers, so you know how much to mix.
  5. Correct anything wrong. No algorithm is right everywhere — paint straight onto the mask, merge shades that are too close, split one doing too much work.
  6. Bring the reference back. Slide towards the original to see the photograph beneath the boundaries, and blend the transitions by eye.

The pipeline

What happens in those few seconds

  1. A working copy. Your original is never altered. The analysis runs on a downscaled copy, then the masks scale back up.
  2. Into Lab. Colours convert to CIE L*a*b*, where distance matches perception.
  3. Noise removed. An edge-preserving filter takes out sensor and JPEG noise that would otherwise split flat areas between two shades.
  4. Shades found. k-means++ groups millions of colours into your chosen number of paint shades.
  5. Edges preserved. A refinement pass balances colour fit against neighbour agreement, discounted across real boundaries.
  6. Specks absorbed. Patches too small to paint are merged into a neighbour — unless their colour is genuinely different, in which case they are kept.
  7. The match measured. The result is compared against your untouched original and reported in CIEDE2000.

The output is a single map naming the shade of every pixel — which is why switching shades, merging, splitting and brushing corrections are all instant.

Try it on a reference you know well.

You will see immediately whether the shades match what your eye already told you.

Open the studio

Free to try · no card · nothing leaves your device