This is an implementation note from the Sparkles Editor team. Here is a practical way to separate cutting errors from animation choices, and prepare a sheet for GIF export without flattening the motion.

Count poses, then inspect boundaries

For a true 4 × 4 sheet, the internal cuts fall at 25%, 50% and 75% of both dimensions. A 1024 × 1024 image therefore has cuts at 256, 512 and 768 pixels.

Having sixteen drawings does not guarantee that the artwork respects those coordinates. Before changing the animation, overlay the cuts on the source image and inspect the widest pose, the highest jump and the largest effect.

Treat these as different cases:

• A head crosses a horizontal cut: the crop needs attention.
• The character disappears while a thrown prop remains: this may be the intended action.
• A frame is empty after a dissolve: it can be a valid part of the timing.

Do not automatically delete empty frames or force a character into every frame. Those rules change the animation rather than repair the sheet.

Use empty gaps as evidence, not a guarantee

A useful first pass is to project foreground pixels onto the horizontal and vertical axes. Columns and rows containing no foreground are candidates for separators.

In Sparkles Editor, transparent pixels help build that mask. For opaque images, detection looks for a dominant flat background connected to the image edges. That background estimate is used to locate gaps; it does not automatically erase pixels from the artwork.

The detector searches near each expected grid boundary for a strip of empty space. If it cannot find a convincing gap, equal divisions remain a starting point and manual controls are still available. A busy background, overlapping effects or a subject extending across cells can defeat automatic detection.

Move a cut only after checking every cell it affects. A horizontal line fixes the boundary for the whole row, so a position that clears one head could cut through another pose. If no safe separator exists, the source may need editing or regeneration.

Keep one scale and a shared canvas

After adjusting the cuts, source cells may have different sizes. Resizing each crop independently to fill the output frame creates another problem: the character changes size while moving.

Use one scale for the sequence. Derive the output canvas from the largest crop dimensions, scale every crop by the same factor, and pad the smaller ones. Then use small per-frame offsets to align a stable reference point where needed.

For a walk cycle, that reference may be the character’s root. For a jump, preserve the vertical arc. Aligning the top of every head would remove the movement you are trying to keep.

Check the result at normal playback speed and frame by frame. The end of one pose and the start of the next matter as much as each individual still.

Review timing before exporting

A uniform frame rate is a useful starting point, but an anticipation pose or landing may need a longer hold. Adjust the hold duration rather than duplicating artwork just to slow one moment down.

Use looping playback only when the ending connects naturally to the beginning. A throw, disappearance or one-way transformation may work better as a single play. Ping-pong playback can suit some motions, but it reverses the action and is not a universal loop repair.

Transparent PNG and transparent GIF behave differently

An RGBA PNG can preserve partially transparent edges, such as soft hair or a fading effect. GIF uses palette-based color and binary transparency: a pixel is either visible or transparent. Soft edges can therefore look different after export.

A checkerboard painted into the source image is not transparency. Inspect the actual result against a contrasting preview background. If the source has a flat white, green or magenta background, explicit color-key removal may help, but it can also remove matching colors inside the character. Check clothing, eyes and props before accepting it.

Choose GIF for an easy-to-share animated preview. Use a PNG atlas with frame metadata when partial alpha and reusable frame data matter.

Try the workflow

The Sparkles Editor sprite-sheet animator lets you import PNG, JPG or WebP, inspect the grid, detect gaps, adjust cut lines, preview timing and offsets, and export GIF or a PNG atlas with JSON frame data. Conversion runs in the browser without an account or Credits. Creating new images with AI is a separate feature that uses Credits.

The current converter accepts 2–64 source frames, with inputs up to 12 MB and 4096 pixels per side. Exported frames are fitted within 320 pixels per side. Those limits make this workflow suitable for compact motion previews; they are worth checking before preparing a large production asset.

The most useful final check is simple: overlay the cuts again, inspect every frame, and confirm that any disappearance, empty space or movement is intentional. A successful export alone cannot tell you that the animation was cut correctly.