1 · Add & verify artwork
Drop artwork (the separation backlog is gold here). Each file is auto-analyzed — background removed and trimmed when detected — then you confirm or correct the ink count. Every verified sample joins the permanent training library and makes the counter smarter. Filenames like name-3color.png or name_4c.png pre-fill the count.
PNG, JPG, GIF, WebP, SVG · PDF & Illustrator (.ai) — vectors are rasterized automatically
2 · Accuracy & learning
Two learning layers, both in-house: auto-tune finds the detector thresholds that best fit your library, and the k-NN model predicts counts from detection features using your verified samples (scored honestly by leave-one-out: each sample predicted as if it weren't in the library). Target: 100% within ±1.
3 · Verified library
Every human-verified sample. Fix a label any time — the model updates instantly.
4 · Team access
Who can sign in to this live Color Lab (admin only). Roles: admin manages the team, staff labels and tunes.