# PixMiniMax character round: Original vs Nik

This round compares the existing **Original** animation prompt enhancer with the developer-only **Nik** enhancer on 20 gallery assets. The source set is made from front-facing or side-scroller style character sprites and enemies with transparent pixels; all 20 source images were checked for an alpha channel and were sent with `no_background: true`.

## Controlled setup

- 20 source images, 20 simple actions, and 40 total Minimax generations.
- The two variants use the same source, action, seed, frame request, and `enhance_prompt: false` animation setting for each case.
- Requested frame counts cycle through 8, 12, 16, 20, and 24. Minimax returned 9, 13, 17, 21, and 25 frames respectively.
- 680 returned frames passed the dimension and count checks. All output frames stayed at the source's native dimensions; no source or output was resized.
- Every Nik prompt ends with the required locked-camera pixel-art sentence. Nik word counts ranged from 94 to 107; 13/20 landed inside the requested 80–100 range.

## What the clips show

- Nik generally keeps the original silhouette and pose more settled while describing weight, planted footing, cloth response, and a clear return to the starting stance.
- Original often gives a more explicit beat-by-beat action plan. It sometimes introduces a larger prop motion or a stronger effect than the short action implies; that makes those clips useful when testing expressive motion, while increasing the chance of silhouette changes.
- On the wolf and mage cases, Original and Nik both preserve the character well across the sampled frames. Nik is visibly more restrained on the no-effect actions and adds the exact camera/style lock to every prompt.
- The wolf rogue case is a clear failure mode to inspect: Nik's mid/end samples darken the sprite heavily during the side turn. The original keeps more of the visible fur detail there.
- Shield, sword, staff, page-turn, greeting, and stride cases make the strongest side-by-side tests because the intended motion is easy to judge against the transparent source.

These observations are visual review notes from the generated frame sheets, not a statistical benchmark. Use the controls on the comparison page to play both sides in sync, scrub the same frame, vote, and leave notes; review data stays in the current browser.