Apply Gradient Color
Turn a bare OBJ into a colored GLB you can actually see
- STRING
The tiny helper that makes generated meshes legible
A raw OBJ has no colors, no textures, no shading - it's a ghost of triangles floating in space, and in a plain viewer it's genuinely hard to tell a blob from a chair. Apply Gradient Color fixes the worst of that in one step: it takes the mesh_text (OBJ) that the pack's Chat LLaMa Mesh node produces, paints a blue-to-red gradient across the vertices based on their height, and exports a GLB file you can open anywhere.
This is a utility node from the ComfyUI-LLaMA-Mesh pack, and it's honest about being one - a single input, a single output, thirty lines of code. Don't expect it to be more than it is: it's the "make the mesh visible enough to judge" step, not a renderer.
How it works
Under the hood it hands the OBJ string to trimesh, grabs the vertex coordinates, and normalizes the Y-axis (up/down) into a 0–1 range. That value drives the red channel, its inverse drives blue, and alpha is set to fully opaque - so the lowest point is blue, the highest is red, and everything between is a ramp of purple. It then writes the colored mesh out as a GLB and returns the file path to you.
That Y-gradient is doing more work than it looks like. It's effectively a cheap heightmap visualization: dips and bumps in your mesh become instantly obvious as color shifts. For the lumpy, low-poly output LLaMA-Mesh actually produces, it's the difference between squinting at a white silhouette and actually seeing your geometry's shape.
The one input and one output
- mesh_text - the OBJ as a string. The multiline field is there precisely so you can paste in whatever
mesh_text(or the response text) from Chat LLaMa Mesh spat out. - Output - a single STRING that is the path to the exported
.glbfile on disk.
There are no optional inputs, no tuning knobs, no color choices. What you get is the path to a temp .glb - and "viewing" it is on you: open it in Blender, Windows' built-in 3D Viewer, or any GLB-capable viewer. ComfyUI itself won't pop a preview, because what you're getting is a file location, not a rendered image.
Installing
Same pack as everything else in here:
cd ComfyUI/custom_nodes
git clone https://github.com/Yuan-ManX/ComfyUI-LLaMA-Mesh.git
cd ComfyUI-LLaMA-Mesh
pip install -r requirements.txt
or search LLaMA-Mesh in ComfyUI Manager. The dependencies that matter for this node are trimesh and numpy, both in the pack's requirements.
Where people get burned
The output is a path, and people habitually assume they'll see something on screen. You won't - grab the returned string and open the file yourself. The node also writes to a fresh temp file each run, so the disk fills quietly if you batch a hundred generations. And since it normalizes by max - min of the Y values, a degenerate mesh where every vertex sits at the same height will throw a divide-by-zero - which is rare with real LLaMA-Mesh output, but possible with something you hand-wrote. Mostly, this node just works, and does exactly the one boring thing it promises.
Inputs (1)
| Name | Type | Default | Description |
|---|---|---|---|
| mesh_text | STRING | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| STRING | STRING | — |