Disclosure: This draft was generated with AI and checked against the cited primary sources, with a separate AI editorial review. It was prepared for the FrameSprite team, which has a commercial interest in FrameSprite. Proposed examples are not executed tests unless explicitly identified as recorded fixture results.
Label all six legs before judging an AI creature's walk. Keep each label attached to the same body joint across the cycle, then compare each leg's intended support frames with its visible motion. A smooth preview can conceal a copied leg phase. Counting six feet, or adding up contact marks, will not catch every error.
Key Takeaways
- Track anatomical left and right, with front, middle and rear leg IDs that survive overlaps.
- Compare each leg at each frame. The supplied eight-frame fixture contains 48 Boolean cells.
- Copying one leg's phase can preserve the total support count while corrupting every frame in that leg's row.
- The fixture checks a deliberately simplified animation contract. It does not establish a natural gait or validate generated artwork.
Written by the FrameSprite Workflow Team. Send a correction or a reproducible failure.
Give every leg an identity that survives overlap
Use L1, L2, L3 for the creature's left front, middle and rear legs, and R1, R2, R3 for its right equivalents. Left and right belong to the creature, not the screen. Start with a pose where you can identify all six attachment points.
Scrub the frames with those labels overlaid. Follow each attachment point through an overlap instead of assigning labels again by horizontal position. A rear leg passing behind a middle leg must retain its original ID.
Record uncertainty explicitly. An occluded foot has unknown contact unless another view or an approved pose establishes it. Do not turn “I cannot see it” into “it is swinging.” Keep unresolved observations outside this Boolean CSV until you can establish contact; do not coerce an unknown observation to 0. If a joint disappears or two limbs merge, resolve that art defect before interpreting its contact schedule.
This addresses the six-legged-creature question raised by BoltVnderhuge in an AI game-development discussion. That comment identifies a practical need; it supplies no benchmark proving that a particular generator handles six legs correctly.
Use a six-leg fixture with a known answer
The downloadable fixture chooses two alternating support groups across eight equally held frames numbered 0–7 in the CSV. Group A is L1/R2/L3; group B is R1/L2/R3. A supports during frames 0–3, and B during frames 4–7.
| Group | Leg IDs | Support frames | Swing frames |
|---|---|---|---|
| A | L1, R2, L3 | 0–3 | 4–7 |
| B | R1, L2, R3 | 4–7 | 0–3 |
Each of the eight frames has six Boolean entries: 1 means intended support and 0 means intended swing. The baseline therefore contains 48 cells, three support entries per frame and 24 support entries overall.
These are authored test rules, not a biological claim. Six-legged animals and fictional creatures can use other coordination patterns. This fixture has an abrupt group change and no transition model, body dynamics or weight distribution. Use it to isolate labeling and phase errors, then replace its schedule with your approved movement reference.
Download the contact ledger CSV, Python checker and fixture result record. Keep the files together so the ledger and its checks remain traceable.
Break one leg's phase and compare individual cells
Copying the L1 contact values into L2 produces a useful failure case. Those legs belong to opposite groups, so the copied row disagrees with the baseline in all eight frames.
| Fixture case | Recorded support per frame | Recorded total | Recorded comparison |
|---|---|---|---|
| Baseline | 3 throughout | 24 | No changed cells |
| L2 contact row copied from L1 | 4 in frames 0–3; 2 in frames 4–7 | 24 | 8 changed cells, all in L2 |
The total remains 24 because four extra support marks in the first half cancel four missing marks in the second. A total-only check would miss this mutation. Per-frame counts expose uneven groups; comparing the full matrix identifies the responsible leg.
The Python checker was executed on 20 October 2026. All six fixture checks passed: the baseline, copied limb track, reversed frame labels, missing limb, duplicate header and invalid contact value. The saved result file records the script SHA-256. No game engine or AI-generated creature was run. Run python check-six-legs.py to reproduce the assertions, or python check-six-legs.py --input copied-limb-ledger.csv to see the deliberate failure. The second command exits with status 1 because its contacts differ from the chosen contract.
Apply the ledger to generated frames
Make an annotated copy of your exported sequence and preserve the untouched original. For each frame, record the leg ID, visible foot position, intended phase and any uncertainty. Match those observations against the reference schedule, rather than against how convincing the loop looks at full speed.
Use the first disagreement to choose the next repair:
- Identity changes: A label jumps to another attachment point. Recheck occlusion or repair merged anatomy.
- Phase changes: Identity stays stable, but a leg follows the wrong support interval. Adjust that limb's poses or regenerate the affected motion.
- Order changes: Individual poses are plausible, but their sequence is wrong. Check the PNG frame-order workflow before changing the art.
A contact ledger records intended support; it cannot prove that a painted foot carries weight. Inspect whether the foot drifts during its support interval and whether the body moves coherently above it. Diagnose movement-speed mismatches separately with the foot-sliding guide.
Keep playback checks separate from anatomy checks
Import the approved sequence with its recorded order and equal frame holds. Godot 4.5's official SpriteFrames reference source describes animation frames, speed and relative frame durations. Verify those settings when the imported cycle differs from your source sequence.
Use Godot 4.5's official 2D sprite animation tutorial source for the editor workflow. Neither documentation page validates a creature's anatomy or intended gait. No Godot project was executed for this fixture.
Retain the annotated frames, contact ledger and actual checker output together. Accept a candidate only after its six identities remain traceable and its contacts match the chosen reference. If you are auditing a four-legged asset instead, use the quadruped contact-phase guide with a reference appropriate to that creature.
Top comments (0)