The early knee-direction error

Lower-leg bones have a different local-axis orientation from upward-pointing bones. Reusing rotation signs produced reversed knees. The diagnosis remains useful; the poses and animation settings are early trials.

The later model and motion are shown in the complete animation.

As of September 9, 2026: creature mod 0.12.1 0.12.2 covers all fourteen Necropolis creatures. Measurements here retain their delivery-version scope; see mounted knights and skeletal dragons for the latest integration.

Tools live in h3-art-pipeline; models and complete mods stay local. Migration and reproduction.

History

Earlier article and revision notes (expand)

The text, images and strikethrough annotations below are preserved from before this editorial pass. “Current”, unfinished work and next steps refer to the time each passage or annotation was written. Old versions, paths and trial renders are historical records.

Status wording from 0.12.0; the current patch version is stated above:

As of September 9, 2026: creature mod 0.12.0 covers all fourteen Necropolis creatures. Measurements here retain their delivery-version scope; see mounted knights and skeletal dragons for the latest integration.

As of September 9, 2026: creature mod 0.11.1 contains ten Necropolis creatures. Installation counts in this article belong to the named delivery version. The latest expansion is Vampire and bat forms.

2026-09-09 tool migration: Development continues in h3-art-pipeline. The old tools/creature-art/ and tools/town-art/ paths correspond to creature-art/ and town-art/ in the new repository. Historical paths and PR links remain in this article. Migration and reproduction details.

Update, 2026-09-09： Images, poses and measurements below are retained as the historical study, not the current installed version. See the later animation and current eight-unit delivery.

With the skeleton finished my validator gave it full marks: thirteen groups, eighty-two frames, zero errors, zero warnings.

Then someone looked at it and said the knees bend the wrong way.

Backward knee, over-bent, corrected, original
Left to right: backward-bending knee, right direction but over-bent, corrected, original. The first one folds its shin forward.

They do. The shin folds forward, like a bird’s leg.

Why

I had measured the rotation axes earlier, on the spine: positive local X tilts it forward. So I carried “X is forward” to the whole rig.

Leg bones point downward. The same local axis points the opposite way in the world on a bone that hangs versus one that rises. On a leg, positive X is backward, not forward.

I had set every lower leg negative, meaning “bend backward” — which folded them forward instead. BASE and all thirteen groups, every lower-leg rotation inverted.

Measured properly this time, watching the foot rather than reasoning:

RightLeg X+40  ->  foot moves back and up    correct knee
RightLeg X-40  ->  foot moves forward        backward knee

Thirty seconds of measurement against an hour of being confidently wrong.

The part worth writing down

After fixing it I scored all three versions on silhouette overlap:

backward knee     36%
right way, over-bent  29%
corrected         33%

The backward-bending version scored highest.

Not mysterious once you look at it: a shin folding forward happens to sit closer to the original’s forward-leaning stance, and bounding-box overlap cannot see anatomy. It measures how much area coincides. A leg bending forward or backward is the same thing to it.

Had I trusted the number, I would have kept the backward knees and treated the two corrections as regressions.

Third time

  • First: the drift threshold. Run against the original art it produced eight warnings on frames that are correct by definition. The thresholds were wrong.
  • Second: silhouette overlap as a concept-stage gate. A free-pose concept against one specific animation frame — facing alone halves it, and a perfect concept still fails.
  • Now: joint direction. The metric gave the wrong answer its highest score.

The pattern each time: the metric measures something adjacent to what I care about, and I didn’t notice.

The first two I caught myself, by running the check against the original art. This one I didn’t. A person looked at it and said the knees were backward.

What my checks catch is structural and formalisable: inconsistent canvases, a misplaced anchor, missing layers, a drifting ground line. “Which way does the knee bend” and “does this read as a skeleton” are not things I have any automated grip on yet.

Code in PR #10.