About

TestMyPaint repaints a photograph of your room in a colour you choose. It is brand-neutral: we sell the decision, not the paint. Your photo is kept only so you can come back to your renders, is never used for training or marketing, and is deleted after 30 days. We may look at uploads and their results to fix problems.

How close is the colour?

Every render is measured. After the model has repainted the walls we sample the painted region, compare it with the colour you asked for in LAB space, and correct the difference — then measure again and show you the result as a ΔE number. Under 5 is a close match; on our validation set of 200 real renders the median after correction is 2.88, against 10.61 for raw model output.

Part of any remaining gap is not an error. A colour chip is measured under standard daylight and your room is not, so a warm-lit room genuinely does make a colour look warmer — and it will do the same to real paint. That is why every result carries the same line: screen colours vary, and you should always confirm with a physical swatch before you buy.

Sample-room photographs

The three rooms you can try without uploading anything are freely licensed photographs from Wikimedia Commons, repainted by the same pipeline your own photo goes through. The repaints are derivative works and are used here under the licences below.

• Bright living room

  “Living room at artchapiz.es apartments” by Artchapiz, licensed CC0 1.0. Source. (Public-domain dedication — no attribution required; credited anyway.)

• Bathroom

  “Cedar Point's Express Hotel bathroom” by Ken Srail, licensed CC BY 2.0. Source.

• Kitchen

  “Interior of a park model mobile home, Dike, Iowa” by Susie Shapira, FEMA, licensed Public domain (US federal government work). Source. (Public-domain dedication — no attribution required; credited anyway.)

Each room is shown in 5 colours, rendered once in advance. Opening one uses no credits and makes no API call.

Privacy

We do not use cookies for analytics and there is no consent banner, because there is nothing to consent to: usage measurement runs in a cookieless mode that stores nothing on your device and builds no profile of you. Sign-in uses one cookie, which is the session itself.

We do look at uploads and their results, on a small internal page that lists recent renders, so that a photograph the model handles badly is found by us rather than reported by you. That is the only reason, it is not automated profiling, and it changes nothing about how long anything is kept: uploads and renders are still deleted after 30 days.