Skip to content

Kali Rigging Attempts Log

Date: 2026-03-03 Source model: https://cdn.cinevva.com/hunyuan3d/kali-tpose/model.glb Rig endpoint: POST https://api.cinevva.com/genai/rig

This file tracks every rigging attempt with exact parameters and outcomes so we can reproduce or audit each run.

Attempt History

AttemptjobIdrotationYbakeWalkAnimationOutput URLLocal fileObserved result
1e4d77502-4a41-4498-86ea-183c2ed83d35-90truehttps://cdn.cinevva.com/rigging/e4d77502-4a41-4498-86ea-183c2ed83d35/model.glbkali-rigged-attempt1-minus90.glbUpright model. Walk cycle appears backward.
2bdf76647-bc41-459d-b94a-96ee1d8f9b6790truehttps://cdn.cinevva.com/rigging/bdf76647-bc41-459d-b94a-96ee1d8f9b67/model.glbkali-rigged-attempt2-plus90.glbModel appears upside down.
35844ded8-1ba4-409b-a4b2-a9ca8d710893not settruehttps://cdn.cinevva.com/rigging/5844ded8-1ba4-409b-a4b2-a9ca8d710893/model.glbkali-rigged-attempt3-zero.glbModel appears upside down and distorted.
4 (repeat of 1)53e0179d-f5d0-405c-93c9-1f430c385c04-90truehttps://cdn.cinevva.com/rigging/53e0179d-f5d0-405c-93c9-1f430c385c04/model.glbkali-rigged-attempt4-repeat-minus90.glbRepeat test for stability.
5 (repeat of 2)f2dfd053-476c-4d7b-a0e2-ed1469abd05c90truehttps://cdn.cinevva.com/rigging/f2dfd053-476c-4d7b-a0e2-ed1469abd05c/model.glbkali-rigged-attempt5-plus90-repeat.glbRepeat test for +90 stability.

Reproducible Request Payloads

Run command template:

bash
export API_ADMIN_KEY=your-key

curl -s -X POST "https://api.cinevva.com/genai/rig" \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $API_ADMIN_KEY" \
  -d @payload.json

curl -s "https://api.cinevva.com/genai/rig/status/<jobId>" \
  -H "Authorization: Bearer $API_ADMIN_KEY"

Attempt 1 payload:

json
{
  "modelUrl": "https://cdn.cinevva.com/hunyuan3d/kali-tpose/model.glb",
  "rigType": "biped",
  "outputFormat": "glb",
  "bakeWalkAnimation": true,
  "rotationY": -90
}

Attempt 2 payload:

json
{
  "modelUrl": "https://cdn.cinevva.com/hunyuan3d/kali-tpose/model.glb",
  "rigType": "biped",
  "outputFormat": "glb",
  "bakeWalkAnimation": true,
  "rotationY": 90
}

Attempt 3 payload:

json
{
  "modelUrl": "https://cdn.cinevva.com/hunyuan3d/kali-tpose/model.glb",
  "rigType": "biped",
  "outputFormat": "glb",
  "bakeWalkAnimation": true
}

Notes from GLB Diff Analysis

  • The source Hunyuan model has one root node with a +90 degree X rotation.
  • All rig outputs use an Armature root plus animated skeleton nodes.
  • Across attempts 1/2/3 (different pre-rig rotations), bone transforms differ significantly, not just by yaw orientation.
  • This means changing pre-rig rotation is not a reliable way to tweak orientation.
  • Repeating the exact attempt-1 payload (attempt 4) is highly stable.
  • Orientation adjustments should be applied after rigging on a known good output.

Controlled Next Step

Use attempt 1 as the canonical rig because its mesh and pose are stable. Do not rerig again for direction fixes. Fix forward direction in a post-rig step by rotating the rigged output at scene root or by adjusting root-motion direction in the animation track.

Stability Check: Attempt 1 vs Attempt 4

Exact same payload and source mesh:

  • Size delta: 52 bytes (30271504 vs 30271556)
  • Counts match exactly: 41 nodes, 1 mesh, 1 skin, 1 animation
  • Node translations: identical (max diff = 0.0)
  • Node scales: identical where present (max diff = 0.0)
  • Node rotations: tiny variance only (max delta = 0.0376 degrees, mean 0.0037 degrees)

Conclusion: with fixed input mesh and fixed payload (rotationY: -90), Tripo output is reproducible enough for production.

Stability Check: Attempt 2 vs Attempt 5

Exact same payload and source mesh:

  • Size delta: -52 bytes (30271408 vs 30271356)
  • Counts match exactly: 41 nodes, 1 mesh, 1 skin, 1 animation
  • Node translations: identical (max diff = 0.0)
  • Node scales: identical where present (max diff = 0.0)
  • Node rotations: tiny variance only (max delta = 0.0414 degrees, mean 0.0011 degrees)

Conclusion: with fixed payload (rotationY: 90), output is also reproducible. So the upside-down issue is consistent with that configuration, not random corruption.