Photo input supports 2×2 and 3×3 cubes with six distinguishable colors, including custom palettes. Similar shades still need review. The 4×4 uses manual entry with standard colors. Picture cubes and shape modifications are outside this tool’s scope.
1. Put the grid on one face
Six photos need a consistent face order and direction. The guided whole-cube turns establish that relationship; turning a layer while capturing changes the state. The four adjustable corners define a perspective projection so each sample belongs to the intended sticker, even when the outline is a trapezoid.
An automatic outline is a suggestion. Refitting the corners resamples that face. It does not restore a sticker hidden by a finger or recover color destroyed by glare. A heavily blurred or occluded face may still need replacing.
2. Use the dominant shade, not a mixed average
The sampler reads multiple points inside each sticker, leaving its border out. It groups nearby shades and uses the group with the most votes. A representative observed color is selected from that group before classification. This helps prevent a small logo, black divider or isolated bright reflection from dominating the answer.
When the samples disagree, the sticker remains marked as uncertain. A dominant group is a useful suggestion, not proof of the physical color. Strong glare over most of a sticker can still make its dominant shade wrong.
3. Learn reference colors for this cube
Standard-color classification considers hue, saturation and brightness. Checked photographs and direct color corrections can provide reference shades reused on the same cube’s other photos. A reference can help distinguish neighboring red and orange shades without asking the user to repaint every occurrence.
For custom palettes, choose six clear, distinct reference colors in Cube colors. A 3×3 can use its centers; a 2×2 needs a sample for each color because it has no centers. Recognition compares samples with these references and flags close or ambiguous matches. This is reference calibration, not a promise to discover every unusual palette automatically.
4. Separate color counts from solvability
Each color must appear four times on a 2×2 or nine times on a 3×3. Matching those totals is necessary but insufficient: swapping two stickers can preserve every total while describing an impossible puzzle. The solver also checks piece identities, corner twists, edge flips and the applicable permutation rules.
The checks respect size differences. A 2×2 does not have edges and does not use the 3×3 corner–edge parity rule. A legal 4×4 can have reduction parity; this is handled by its separate manual solver. A warning is a reason to inspect the entry before assuming the physical cube is broken.
5. Correct the mistake and keep the rest
Compare the model with your real cube. Select the correct color and repaint only a wrong sticker in the 3D editor or six-face grid. Corrections update the shared workspace and can teach a trusted shade when the photo sample is suitable. Counts alone never choose a replacement color for you.
If an entire face was recorded sideways, rotate its entered colors. If the crop is wrong, refit just that outline; if the image itself is unusable, replace only that photo. The other faces remain available. Undo recovers a recent edit while the page is open.
6. Verify the returned moves
Recognition, correction, validation and solving run in your browser. Once the entry is legal, a browser worker calculates a solution and applies it to the entered sticker state to verify the finish before displaying the moves. Photos are not sent to an external recognition service.
That verification covers the entered state. It cannot confirm that you copied every physical sticker or performed each turn correctly. Match the initial view, follow one turn at a time and use the recovery controls if your cube stops matching. At the end, check all six physical faces.
Try correcting one sticker
No recognition-rate or fastest-scan claim is made here. This explanation describes the implemented workflow; broad accuracy comparisons require real tests under comparable lighting, cameras and cube colors.