The claim we can defend
Keep more of the composition still.
Change a panel’s color or position. Preserve a declared color balance, protected artwork and alignment rules. Find the fewest other layers that have to move—and verify the delivered Photoshop pixels.
5 wins · 26 tiesFewer collateral layers vs. the strongest
minimum-distance comparator, on 31 feasible revisions.
31 / 48Requested revisions delivered.
17 incompatible requests refused.
47 vs. 52Collateral layer changes across matched revisions.
9.6% fewer; 2.9% more total movement.
≤ 0.25 pxChromatic-centroid error, per axis.
All accepted output pixels verified.
A bounded result on three six-panel compositions, not a general image-editing benchmark. “Color balance” here is raw OKLab chromatic centroid—not eye attention or aesthetic quality. No human time-saving claim.
Inspect the actual outputs
Collateral position changedExplicitly requested moveDashed outline = previous position · F is protected
What this revision changed
Why this comparison is stronger than the first result
The comparator minimizes total pixel displacement, then minimizes the number of moved layers among all equally good answers. It receives exactly the same prior state, constraints, measured color variants and destination requests as our solver. The initial comparator without that tie-break gave us 12 apparent wins. Adding the tie-break reduced that to five. The displayed claim uses the stronger comparison.
Our objective reverses those priorities: first disturb the fewest other layer positions, then minimize displacement. Both use real integer optimization. This is a demonstrated preference tradeoff and integration result, not a new optimization algorithm. On the 31 feasible transitions our total displacement is 2,491 px versus 2,420 px.
Coverage, setup and prior art
All three compositions are fixed permutations of the same six existing library artworks. They share 24 prepared color variants. There are no cutouts, inferred backgrounds or generated content. Native groups and locks, declared slots, two row alignments, explicit destinations and one measured color-balance target define the editable system. Setup and human correction time have not been measured.
Direct linked-row edits passed 0/48 balance gates. Actual upstream Kiwi/Cassowary plus exhaustive floor/ceil rounding passed 22/48 on its own edit history, or 28/48 on our previous states. An integer distance optimizer matches our 31/48 coverage. This does not establish Photoshop’s capability ceiling or an advantage over an expert.
Cassowary constraint maintenance, selective undo and Photoshop scripting are established. This experiment assembles measured color constraints, minimum disturbance and native-file verification. Releasing a destination is not general bitmap undo.