Nodes/BrainDead Nodes/BD Cache Trellis2 Shape
ComfyUI Node

BD Cache Trellis2 Shape

The shape stage is 30-60 seconds you should only pay for once

By BizaNator·Created 8 months ago·Updated 3 days ago· 15
BD Cache Trellis2 Shape
  • shape_result
  • mesh
  • shape_result
  • mesh
  • status
cache_nametrellis2_shape
seed0
force_refreshfalse
name_prefix

This is the node the pack's author clearly got tired of waiting on, because the description is basically a frustrated developer talking to you: "Cache Trellis2 shape result + mesh to skip expensive generation. Place AFTER Trellis2ImageToShape node. Saves ~30-60s per run!" That exclamation mark is earned - the shape stage of TRELLIS2 is the slow heart of the whole image-to-3D pipeline, the rectified-flow transformer actually generating the geometry. If you're iterating on texture, baking, or export, paying 30-60 seconds of shape generation on every single run is pure waste. This node ends it.

It caches two things together: the TRELLIS2_SHAPE_RESULT (the raw internal shape output) and the derived TRIMESH mesh. Both come back on a hit, so downstream nodes that want either one work unchanged. And like the whole family, it's lazy: on a cache hit, check_lazy_status declares it needs no inputs, and ComfyUI skips Trellis2ImageToShape and everything upstream - which, if you've also placed the conditioning cache, means the image barely gets touched on re-runs.

The inputs

  • shape_result - the TRELLIS2_SHAPE_RESULT from Trellis2ImageToShape.
  • mesh - the TRIMESH derived from it (the same node usually outputs it).
  • cache_name - label (default trellis2_shape). Name it after the asset - base_character_shape.
  • seed - invalidation key. Change it to force a fresh shape generation.
  • force_refresh - True to overwrite and regenerate.
  • name_prefix - optional path prefix under output/BrainDead_Cache/.

Outputs: shape_result, mesh, and status.

The workflow shape

[Trellis2ImageToShape] → [BD Cache Trellis2 Shape] → [Trellis2ShapeToTexturedMesh]
                          cache_name: "shape_01"
                          seed: 42

First run generates the shape and writes it. Every run after that skips the generation and hands you the identical shape result and mesh. Pair it with the conditioning cache upstream and the texture cache downstream, and you can iterate on any single stage of the three-stage pipeline without paying for the others - that's the workflow the pack is built around.

Installing it

Part of BizaNator/ComfyUI-BrainDead (BrainDeadGuild). ComfyUI Manager → search "BrainDead" → install, or:

cd ComfyUI/custom_nodes
git clone https://github.com/BizaNator/ComfyUI-BrainDead
cd ComfyUI-BrainDead
pip install -r requirements.txt

Restart; it's under 🧠BrainDead/TRELLIS2.

Gotchas

You know the rules by now - seed-keyed, so change the source image or the shape parameters and bump the cache seed. One thing worth flagging: because the shape result is an internal TRELLIS2 object, its pickle/serialization is tied to the TRELLIS2 pack's version. Update TRELLIS2 and your cached shapes may not load - if you get a load error after an update, that's the cause, and a force_refresh run fixes it. Also, "skips upstream entirely" means exactly that: if you're debugging the image side, remember the cache is hiding it from you, and that's on purpose.

Category🧠BrainDead/TRELLIS2

Inputs (6)

NameTypeDefaultDescription
shape_resultTRELLIS2_SHAPE_RESULT
meshTRIMESH
cache_nameSTRINGtrellis2_shape
seedINT00–18446744073709550000
force_refreshBOOLEANfalse
name_prefixoptSTRING

Outputs (3)

NameTypeDescription
shape_resultTRELLIS2_SHAPE_RESULT
meshTRIMESH
statusSTRING