Open CubeAssembler. Scanning, review, assembly, and ZIP import/export run in the browser. Choose a cube size from 2×2 through 7×7 before taking photos.
flowchart TD
Setup[Choose cube and colors] --> Faces[Capture or upload faces]
Faces --> Calibrate[Recalibrate colors]
Calibrate --> Review{Review colors?}
Review -- Yes --> Edit[Correct stickers]
Review -- No --> Arrange[Propose assembly]
Edit --> Arrange
Arrange --> Ambiguous{More than one fit?}
Ambiguous -- Yes --> Choose[Choose each side]
Ambiguous -- No --> Confirm[Approve cube]
Choose --> Confirm
Confirm --> Net[2D net and 3D view]
- Choose a Cube (which sets its size) and Colors. Automatic compares built-in and saved color profiles as faces arrive; you can also select one.
- Open the camera and show one face near the dashed guide. Detect face finds and straightens it. If that fails, choose Guide grid to read the fixed square on screen. Neither mode downloads a model.
- Capture four sides while turning the cube, then its top and bottom. The guided order is three quarter turns sideways for the sides, then step 4 tips the cube away from you (it shows D) and step 5 keeps tipping the same way, past one more side, to the opposite face (it shows U). The animations never change the turning direction. Without Mirror this shows the faces F R B L D U. Mirror flips only the camera picture; the arrows always point the same way on screen. With Mirror the physical sideways turn is therefore the other way round, so the side faces come in the opposite physical order. Any order is still accepted. Auto capture waits for five matching good readings. A matching earlier pattern warns you but does not prevent capture: even cubes can have identical faces.
- Review the six faces. The app recalibrates colors across all six photos. Edit colors lets you compare each detected face with its photo and fix a sticker. If the assembled state is valid and every reading is confident, the app can take you straight to assembly approval.
- Approve the proposed cube. If several arrangements fit, Choose each side asks which face belongs where. The unfolded net shows U above the L–F–R–B row and D below it; the 3D viewer shows the same state.
The first two adjacent odd-cube faces keep their photographed slots. If the second is opposite the first, it goes to slot 3 while slot 2 waits for an adjacent face. Later odd-cube faces can be placed by their fixed center colors even when photographed out of order. Even cubes have no fixed center, so they stay in capture order. The net always shows captured colors, even with Mirror enabled. Changing size after a capture asks before clearing saved faces.
First-party cookies remember the selected cube size and color profile, Mirror, Hands-free capture, capture sound, and the 2D Net / 3D View choice. Within 3D View, Stickerless / Stickered and Auto-rotate / Pause are remembered too. The gesture hint hides 10 seconds after the first touch and returns after 20 seconds without touch. It turns off after 30 minutes of active play; Show 3D help in Settings can turn it back on. Whole-cube x/y/z gestures also count as touching the cube. The 3D view starts locked: a drag or swipe only orbits the view. Tap Move layers in the 3D toolbar (or turn on Move layers in Settings) to unlock layer turns; the choice is remembered. While locked, a short hint explains this when you drag a sticker. Once unlocked, swipe a sticker in the 3D view to turn its row or column, including inner slices on larger cubes. The layer follows your finger: let go early and it springs back, keep dragging for two or three quarter turns, or drag back to turn the other way. A quick flick finishes one turn, and a turn that reverses the previous one removes it from the move list. Hold Shift while swiping to turn a standard wide move such as Rw or 3Rw. With Turn sound on, a tone marks the wide turn.
Wide turns near gaps is off by default under Settings → Advanced settings. The gaps themselves do not start turns; if a press starts in a gap, drag into a sticker to begin a swipe. Enable the setting to select wide blocks by starting near a sticker's edge beside a gap and leaning toward the side to include: both layers at the line and every layer on that side turn. For example, on a 5×5, start beside the line between the third and fourth U rows and swipe left, leaning up for 4Uw or down for 3Dw'. The chosen block lights up and follows your lean while your finger stays on the starting face. Once it leaves that face, the side is fixed. A swipe straight beside the line still turns a wide block; reversing the turn direction does not change which side is selected. A swipe from the middle of a sticker turns one layer. A new swipe while the last turn settles finishes that turn and starts the next. With a mouse, drag the background to rotate the view; it keeps turning briefly after release. On a touch screen, one finger turns a layer, or a wide block beside a line between layers when enabled. Two fingers on the cube turn the whole cube, recorded as x, y or z, and two fingers beside it rotate the view without recording anything; pinch to zoom. A touchpad works the same way: a two-finger swipe over the cube turns it (the touchpad's coasting after your fingers lift adds nothing), one beside it rotates the view, and a pinch zooms, while a mouse wheel still zooms. The X, Y and Z arrows in the corner point through R, U and F. Their colors come from an invisible reference frame rather than the current stickers. On 2×2, 4×4 and 6×6 cubes, where no center fixes an absolute direction, the frame starts with the app's chosen U/R/F convention. Face, inner and wide turns leave it alone; x/y/z turns rotate it. On odd cubes, moves containing the fixed middle slice also rotate the frame, preserving the gizmo's middle-slice behavior. Undo and Reset restore the corresponding frame. Turns snap into place like a magnetic cube: while you drag a layer or the whole cube, each quarter turn holds it like a magnet until you pull it free, and when you let go it is sucked onto the nearest quarter turn, snapping a little past it and settling back. They click softly when capture Sound is on. Layer turns update the 2D net, facelet notation, and Orbit64 token. The move list records completed turns; Undo reverses the last one, and Reset returns to the captured state. Scramble uses more moves for larger cubes and turns inner layers on 4×4–7×7, while leaving the fixed middle layer alone on odd cubes. Random State mixes the cube at once into a random valid state, as speed cubers do for practice: the cube spins once and its stickers change at the midpoint. On 4×4 and larger, centers and wings are mixed too. It leaves no move history (the move list shows None, Undo is off), and Reset returns to the starting cube. It is only in the 3D view. The face buttons show one side head-on; Isometric and Iso-back show the cube from its front corner or the opposite back corner. With the view focused, the arrow keys tilt and turn it. Auto-rotate pauses during a drag, a view button, or an arrow key and resumes about 1.5 seconds later. These interactions do not change the saved setting. Captured photos and cube states are not stored in cookies.
Upload files accepts six cropped face images or full camera photos, with or
without meta.json, either selected together or packed in a ZIP. Without
metadata, it shows an order and framing preview: rearrange the faces and choose
Auto, Cropped face, or Full photo for each image. Files named face-u.jpg
through face-b.jpg start in capture-slot order. With metadata, it loads the
saved colors as a fixture. The chosen cube size applies to ordinary photos.
Full photos use the selected Detect face or Guide grid mode; cropped
face images are read directly. In Guide grid mode, select Cropped face for
an image that is already cropped but does not use a face-*.jpg filename.
One picture of a cube net works too: a screenshot or export of the cross (U above F, L F R B in a row, D below) of a 2×2 to 7×7 cube, on a dark or light background, with or without gaps, lines or face titles. The app finds the stickers and the size, cuts the picture into six face images and goes on as with six photos, so you see the order preview and the cube check as usual. Odd cubes are read by their centers, so any color scheme works; even cubes need the standard colors (white, orange, green, red, blue, yellow).
You can also paste (Ctrl/Cmd+V) or drop files on the page instead of
using the button: a net picture, six photos, a fixture ZIP, or meta.json with
its six photos (copy the files in your file manager, then paste). Facelets
pasted outside a field open Type colors with the text ready to apply.
After confirming a cube, Save as test fixture downloads its six photos and
reviewed colors as a ZIP. With npm run fixture:server running on the same
computer, Upload to localhost saves it straight into test/fixtures/, from
the dev server or the published app. Without confirmed photos the button is
there as well: it saves rendered photos of the cube as shown, including layers
turned in the 3D view, in a lighting chosen in the dialog. See the
fixture guide.
Type colors reads the facelets as typed; the format is detected:
- Face letters (URF): six space-separated blocks of N² letters
(U/R/F/D/L/B) in U R F D L B order, each the face whose solved color the
sticker matches. A solved 3×3 starts
UUUUUUUUU RRRRRRRRR. - Colors (WRG): the same structure using color letters (W/O/G/R/B/Y). A
solved 3×3 starts
WWWWWWWWW RRRRRRRRR.
Typing facelets never changes the format the notation panel shows: that is the Colors (WRG) / Faces (URF) switch on the panel (faces by default), also set on the settings page. Input made only of R and B letters fits both formats and is read in that saved format. The notation panel can also copy an Orbit64.Net compact state token for 2×2–7×7. Paste facelets or a state token into Type colors; move and algorithm tokens are separate formats and are not accepted.