/threejs-3d-generator

name: threejs-3d-generator
description: "Generate, texture, rig, animate, stylize, convert, and download 3D assets for Three.js games via the Tripo API. Use for text-to-3D, image-to-3D, game-ready GLB/FBX, characters, creatures, buildings, props, weapons, terrain, auto-rigging, animation retargeting, model texturing, voxel/LEGO stylization, and low-poly conversion. Pair with threejs-image-generator for concept and texture references first."

Three.js 3D Generator

Production 3D assets for browser games, prepared for Three.js. Provider: Tripo.

Resolve `<this-skill-dir>` from the actual loaded skill file. Resolve sibling skills beside it first, then use the runner's discovered paths. Do not mix installed versions or assume a particular home directory.

References

| File | Read it when | | --- | --- | | `references/api-notes.md` | endpoint and task decisions, model versions, polling, postprocess, conversion, rigging, animation, downloads | | `references/threejs-integration.md` | importing outputs into a browser game, GLB/FBX loading, root motion, animation wiring | | `references/image-generator-workflows.md` | pairing `threejs-image-generator` for concepts, textures, UI art, or image-to-3D inputs |

API key

The script reads `--api-key` or `TRIPO_API_KEY`. Keys never go in skill files, game code, or reports.

python3 <this-skill-dir>/scripts/threejs_3d_asset.py probe   # TRIPO_API_KEY=SET|MISSING

Keys defined only in a shell profile can be absent from the process env. If the plain probe unexpectedly prints MISSING, use `threejs-game-director/scripts/probe_asset_credentials.sh`, which sources the profile and probes all three providers at once.

Download URLs expire quickly — download immediately after a task succeeds.

Commands

python3 <this-skill-dir>/scripts/threejs_3d_asset.py --help

Text to 3D, the default for a premium hero model:

python3 <this-skill-dir>/scripts/threejs_3d_asset.py text \
  --prompt "game-ready sci-fi hover bike, sleek armored panels, strong readable silhouette, layered hard-surface detail, PBR materials, clean topology, centered pivot, front facing, no text" \
  --model-version v3.1-20260211 --texture-quality detailed --geometry-quality detailed \
  --checkpoint artifacts/hover-bike-job.json \
  --wait --download --out-dir assets/models/hover-bike

Image to 3D from a generated concept:

python3 <this-skill-dir>/scripts/threejs_3d_asset.py image \
  --image assets/concepts/hover-bike-front.png --model-version v3.1-20260211 \
  --enable-image-autofix --texture-alignment original_image --texture-quality detailed \
  --wait --download --out-dir assets/models/hover-bike

Status, download, and postprocess (`texture_model`, `animate_prerigcheck`, `animate_rig`, `animate_retarget`, `conversion`, `stylize_model`):

python3 <this-skill-dir>/scripts/threejs_3d_asset.py status TASK_ID
python3 <this-skill-dir>/scripts/threejs_3d_asset.py download TASK_ID --out-dir assets/models
python3 <this-skill-dir>/scripts/threejs_3d_asset.py postprocess --type conversion \
  --original-task-id TASK_ID --format GLTF --face-limit 20000 --wait --download --out-dir assets/models/gltf

Animated character pipeline: generation, prerigcheck, validated rig with bounded retries, retargets, and downloads, routed by body plan. Use checkpoints and stop between stages to inspect before spending on dependent work:

python3 <this-skill-dir>/scripts/threejs_3d_asset.py character-pipeline \
  --prompt "stylized cyber runner character, T-pose, full body, game-ready outfit, readable silhouette" \
  --animations preset:idle,preset:walk,preset:run,preset:jump \
  --checkpoint artifacts/cyber-runner-job.json --stop-after model \
  --out-dir assets/models/cyber-runner

# After inspecting the downloaded model/preview:
python3 <this-skill-dir>/scripts/threejs_3d_asset.py resume artifacts/cyber-runner-job.json --stop-after rig
# After inspecting the validated rig:
python3 <this-skill-dir>/scripts/threejs_3d_asset.py resume artifacts/cyber-runner-job.json --stop-after animations

python3 <this-skill-dir>/scripts/threejs_3d_asset.py character-pipeline \
  --prompt "stylized wolf, quadrupedal stance, all four legs planted and separated, full body" \
  --rig-type quadruped --animations preset:quadruped:walk \
  --checkpoint artifacts/wolf-job.json --stop-after model --out-dir assets/models/wolf

Resuming and Recovery

`--checkpoint PATH` is optional on `text`, `image`, `postprocess`, and `character-pipeline`. It records accepted task IDs immediately, stage status, and downloaded file fingerprints; no API keys or signed output URLs go in the checkpoint. Use a separate checkpoint per job. Existing checkpoints must be resumed, not overwritten, and concurrent use is locked.

For background single-task generation omit `--wait`, retain the printed task ID/checkpoint, and run `resume CHECKPOINT` later. Single-task resume waits/downloads that task only; it does not add rigging. Character resume reuses completed stages and continues through animations unless `--stop-after model|rig|animations` limits this invocation. Credentials still come from the current environment. The checkpoint records absolute local paths, so keep its referenced files in place.

The helper retries safe status/download reads with bounded backoff, never paid task submissions. Missing credentials, exhausted credits, invalid input, transient errors, and uncertain submissions are reported distinctly. On interruption, resume the existing job rather than starting over. If a POST may have succeeded but no task ID was received, find it in provider history and use `resume CHECKPOINT --task-id RECOVERED_ID`; do not invent an ID or submit a replacement blindly. Without a recoverable ID, report the uncertainty before any potentially duplicate charge.

For coordinated games follow the director's `references/asset-recovery.md`; continue independent implementation while gener