Convert a single image into a 3D mesh in one pass, using the Trellis.2 GGUF model by Aero-Ex.
The workflow above is the recommended starting point. It chains the two nodes back-to-back:
- Generate Mesh — takes an image and produces a geometry-only GLB
- Texture Mesh — takes the same image and the generated mesh, and bakes textures onto it
| Parameter | Description | Default |
|---|---|---|
image |
Input image (any background — it will be removed) | — |
pipeline |
Quality preset: fast (512), balanced (1024), high (1024 cascade), ultra (1536 cascade) |
balanced |
quantization |
GGUF quant level (Q4_K_M → Q8_0). Lower = less VRAM, less quality | Q5_K_M |
ss_steps |
Sparse-structure diffusion steps | 25 |
slat_steps |
Shape SLaT diffusion steps | 25 |
foreground_ratio |
How much of the frame the subject should fill (0–1) | 0.85 |
remesh_resolution |
Dual-contouring grid resolution for the output mesh | 256 |
seed |
Reproducibility seed | 0 |
Output: geometry GLB (no texture)
| Parameter | Description | Default |
|---|---|---|
image |
Same input image used for generation | — |
mesh |
GLB mesh to texture (from Generate Mesh) | — |
texture_resolution |
Texture map size in pixels | 1024 |
atlas_size |
UV atlas packing resolution | 2048 |
texture_steps |
Texture diffusion steps | 12 |
guidance_strength |
CFG guidance for texture diffusion | 3.0 |
foreground_ratio |
Should match the value used in Generate Mesh | 0.85 |
seed |
Reproducibility seed | 0 |
Output: textured GLB
Each node downloads only what it uses:
| Node | Download |
|---|---|
| Generate Mesh | ~2.3 GB shared files + one quantization: Q4_K_M 2.4 GB · Q5_K_M 2.9 GB (default) · Q6_K 3.4 GB · Q8_0 4.3 GB |
| Texture Mesh | ~5.2 GB |
Installing both nodes with the default quantization takes about 10.4 GB. Other quantizations can be added or removed one by one from the extension's panel on the Extensions page (Modly 0.4.3+).
Weights: Aero-Ex/Trellis2-GGUF on HuggingFace