About

Why this exists

Most people building a small game can write the code and cannot draw the character. Generating individual frames with AI has been possible for a while. Getting those frames into an engine has not: each frame is generated independently, so the character lands in a slightly different position every time, and played back at twelve frames a second that reads as shaking.

The usual answer is to open the frames in an editor and nudge them into place by hand. AI Sprite Animator exists because that step can be done properly by a machine, and because a sprite sheet without its atlas metadata is only half a deliverable.

How the alignment works

This is the part we actually built, so it is worth being specific about.

1. Background removal. The flat background the generators bake in is keyed out, then connected-component analysis removes the stray pixels that keying leaves behind. Detached parts of the character — a raised staff, a sword tip — are kept by proximity and relative size, rather than being deleted as noise.
2. Alignment. Each frame is aligned against the first using phase correlation on the upper body only. The legs are excluded on purpose: during a walk cycle they are supposed to move, and including them drags the alignment and reintroduces the shake.
3. Identity lock. The reference frame's upper body is composited across the set through a feathered seam, so the character's face and outfit stay identical while the motion plays out below.
4. Packing. Frames are padded to one uniform grid and packed, and the atlas is written for the engine you are targeting.

None of this is a model. It is ordinary signal processing, and it runs in your browser rather than on our servers.

Where the AI is, and where it is not

We think you should know which parts of a tool are generated.

• Generated by AI: the base character image, when you ask for one, and the in-between animation frames. We use third-party image and sprite-animation models for both.
• Not AI: everything after that. Background removal, alignment, the identity lock, sheet packing and atlas generation are deterministic code. Running the same frames twice gives byte-identical output.
• Your review is in the middle. Every frame is shown to you before anything is packed, and you decide which ones survive. The tool does not quietly drop frames it thinks are bad.

What this tool does not do

• It does not rig skeletons or edit bones.
• It does not generate audio, cutscenes or effects.
• It does not turn a static image into video. Games need discrete frames on a grid, and that is what it produces.
• It does not claim the frames will be perfect. Generated animation is still generated; the review step exists because some frames will be wrong.

Contact

Signed-in users can open a support request from inside the app, and every message is read by the person who builds the tool rather than a queue.

If you found a frame the alignment handles badly, that is the most useful thing you can send — a case that breaks it is worth more than a bug report that describes it.