Forecast TRELLIS.2 flow velocity on skipped DiT steps

Training-free acceleration for TRELLIS.2 image-to-3D in ComfyUI. It forecasts the flow-matching velocity on skipped DiT steps instead of running the transformer, across the sparse-structure and shape-SLaT stages via the hicache-pp library.

cd ComfyUI/custom_nodes
git clone https://github.com/Archerkattri/ComfyUI-TRELLIS2-HiCache
pip install "hicache-pp>=1.2.1"

Evidence

Historical speedup, TRELLIS.2-4B 512 pipeline, interval=2	1.9x on an earlier RTX 5090 run (README-reported, not reproduced by the CPU gate)
Chamfer distance vs stock mesh vertices	0.0107 symmetric mean nearest-neighbour distance (near-lossless)
Sparse-structure / shape-SLaT skipped DiT calls	9 / 9 of ~22 DiT calls each at interval=2
CPU contract suite	22 tests passing (dummy-DiT patch logic, no GPU)
TRELLIS.2 flow DiTs covered by stages selection	five (sparse_structure, shape_slat 512/1024, tex_slat 512/1024)
CI verification on release v0.1.4	green CPU unit tests + static checks on Python 3.10-3.12

Media

TRELLIS.2 HiCache runtime path — Runtime path: acceleration confined to supported shape stages; texture stays explicit (docs/flow.svg).

Node pack icon, also the README hero image.

Install & verify

python -m pytest tests/test_patch.py tests/test_budget_bridge.py -q
# expect 22 passed (CPU contract suite); tests/validate_gpu.py needs a GPU + TRELLIS.2-4B and is not run here

Limits

What this release does not claim (from the release notes):

In-ComfyUI GPU validation (no GPU hardware).

Family

Runs on	ComfyUI + ComfyUI-Trellis2
Sibling nodes	ComfyUI-HiCache · ComfyUI-TRELLIS-HiCache
Standalone counterparts	hermit-trellis2-plus-plus
Forecast core	hicache-plus-plus (PyPI)
External	ComfyUI docs