solve.visionBeta · Free preview VIDEO → MOVESNew reconstruction

Reconstruct your solvesfrom video.

VIDEO → MOVESUpload a solve

Xuanyi Geng 3.84 average

Former world-record average · Beijing Winter 2026

5 of 5 solves exact
Video byCube RootWatch original
Reconstruction
SOLVER’S PERSPECTIVE
—Loading replay
MovesClick a move to jump to it
Loading reconstruction…TIMING ±100 MS

Example report

Solve 01 · Xuanyi Geng

Loading the example report…

Real reconstruction data · Recognition time is an estimateGet your own report
Source & notes

Xuanyi Geng · Beijing Winter 2026

Known scrambles. All five move sequences match the published reconstructions. Timing: approximately ±100 ms. Cube rotations and solver viewpoint are estimated; orientation is held where tracking is missing.

Preparing a video

Upload tools

Choose the part to reconstruct

A clip can include setup, inspection, and time after the solve. Mark the first and last turn to tell solve.vision where to look.

Optionally mask extra cubes

Spare cubes on the table can distract the detector. Draw a box over them and check that your solving cube stays outside it.

Use your original recording

Keep the first and last moves in frame. 50 or 60 fps helps with fast solves; top-down and front views both work when the stickers are visible.

Upload a solve

With or without the scramble

Known scramble

Paste the scramble

If you have it in your timer or on a scramble sheet, use it. This gives the reconstruction a known starting state and is the more reliable option.

Detect from inspection

Keep the inspection in the video

Select where inspection begins. The decoder can try to recover the starting state from clear views of the cube before the first turn.

This needs enough inspection footage. Reliability depends on how clearly the cube is visible and how it rotates.

A few practical questions

Do I need a smart cube?

No. The reconstruction uses the video of a regular 3×3.

Does the cube need to fill the frame?

No—the example above uses a wider competition shot. What matters is whether the stickers are readable. Hands hiding the cube, especially from the side, make reconstruction harder.

Will it work with every video?

No. Reconstruction can fail or return an incomplete result, even with a video that looks clear. Camera angle, motion blur, lighting, and hands hiding the stickers all affect the result. The five examples above are successful reconstructions, not a guarantee for other recordings.

Can I upload a solve now?

Yes. Upload your video, mark the first and last turn, optionally mask extra cubes, and submit it for reconstruction. When processing finishes, you can view the result in your solve report.