☕ One coffee = one happy creator: https://buymeacoffee.com/the_frizzy1
One image in, a textured .glb out. Pixal3D or TRELLIS.2, picked from one MODE dropdown. It is the official Pixal3D + TRELLIS.2 template in one graph: background removal, MoGe camera, shape, 1536 upsample, remesh, texture bake. No video on this workflow yet — the one above is how to use my workflows, if you are new to them.
What you need
✔ Custom nodes — None. Core ComfyUI only.
✔ ComfyUI version — 0.37.4 or newer
✔ VRAM — 12 GB — tested. Peaks were 10.5 GB (Pixal3D) and 10.6 GB (TRELLIS.2).
✔ System RAM — 8.2 GB peak (Pixal3D) · 8.3 GB peak (TRELLIS.2)
✔ Tested on — RTX 3060 12 GB · 32 GB DDR4 · ComfyUI 0.37.4 (Sage attention)
✔ Licence — Workflow MIT · Pixal3D and TRELLIS.2: MIT
Models to download
Click a file name to download it. Put it in the folder shown, inside ComfyUI/models/.
Pixal3D mode — Comfy-Org/Pixal3D. The last three are used by both modes.
diffusion_models/
File: pixal3d_int8_convrot.safetensors — 5.6 GB
Download:
https://huggingface.co/Comfy-Org/Pixal3D/resolve/main/diffusion_models/pixal3d_int8_convrot.safetensors
Put it here:
ComfyUI/models/diffusion_models/
clip_vision/
File: dino_v3_L_naf_fp32.safetensors — 1.2 GB
Download:
https://huggingface.co/Comfy-Org/Pixal3D/resolve/main/clip_vision/dino_v3_L_naf_fp32.safetensors
Put it here:
ComfyUI/models/clip_vision/
vae/
File: trellis_2_shape_vae_bf16.safetensors — 1.1 GB
Download:
https://huggingface.co/Comfy-Org/Pixal3D/resolve/main/vae/trellis_2_shape_vae_bf16.safetensors
Put it here:
ComfyUI/models/vae/
File: trellis_2_texture_vae_bf16.safetensors — 0.9 GB
Download:
https://huggingface.co/Comfy-Org/Pixal3D/resolve/main/vae/trellis_2_texture_vae_bf16.safetensors
Put it here:
ComfyUI/models/vae/
TRELLIS.2 mode — Comfy-Org/TRELLIS.2
diffusion_models/
File: trellis_2_int8_convrot.safetensors — 5.3 GB
Download:
https://huggingface.co/Comfy-Org/TRELLIS.2/resolve/main/diffusion_models/trellis_2_int8_convrot.safetensors
Put it here:
ComfyUI/models/diffusion_models/
Both modes also need these two
background_removal/
File: birefnet.safetensors — 0.4 GB
Download:
https://huggingface.co/Comfy-Org/BiRefNet/resolve/main/background_removal/birefnet.safetensors
Put it here:
ComfyUI/models/background_removal/
geometry_estimation/
File: moge_2_vitl_normal_fp16.safetensors — 0.7 GB
Download:
https://huggingface.co/Comfy-Org/MoGe/resolve/main/geometry_estimation/moge_2_vitl_normal_fp16.safetensors
Put it here:
ComfyUI/models/geometry_estimation/
downloads.md gives no total, so none is quoted here.
One command instead
python scripts/frizzy.py doctor 3d/image-to-3d-ultimate-12gb --comfy "C:/path/to/ComfyUI"Speed on my RTX 3060
Each mode was run once through this exact workflow. Wall is queue to finished including model loading; sampling is time in the sampler nodes. The README's Size column is empty, so it is empty here.
Pixal3D — 3 min 51 s wall, 1 min 10 s sampling, 10.5 GB peak vram, 8.2 GB peak ram
TRELLIS.2 — 5 min 33 s wall, 2 min 23 s sampling, 10.6 GB peak vram, 8.3 GB peak ram
One run each. That is not a benchmark, and the README does not say which image was used.
What to watch out for
TRELLIS.2 spends about a minute and a quarter longer in the sampler than Pixal3D for the same job: 2 min 23 s against 1 min 10 s.
10.5 GB peak means the card is nearly full. 12 GB is the floor, not a comfortable margin.
If your models sit in sub-folders like
diffusion_models/LTX-2.5/, the loaders stay red until you re-pick each one once.The gallery shows both: a turntable plus three still views of the Pixal3D mesh, and the same four for TRELLIS.2, all rendered from the .glb files the workflow wrote.
Get the workflow
GitHub: https://github.com/Thefrizzy1/The_frizzy1-Workflows/tree/main/workflows/3d/image-to-3d-ultimate-12gb
Hugging Face mirror: https://huggingface.co/The-frizzy1/The_frizzy1-Workflows/tree/main/workflows/3d/image-to-3d-ultimate-12gb
More videos on this stuff: https://www.youtube.com/playlist?list=UUGKDBMhTGqvlBZE9cF5d50Q
