Free & local ComfyUI workflows that turn a single image into a watertight 3D model – and into a game-ready low-poly asset with baked textures – using Pixal3D / TRELLIS.2. Plus a one-click installer for Windows.
▶ Tutorial and more: PixelArtistry on YouTube
Important
Before you start: set up ComfyUI + TRELLIS.2 first with the TRELLIS.2 Installation Guide. This installer builds on that setup.
| File | What it does |
|---|---|
watertightMeshes_win_installer.bat |
Installs all node packs, the workflows and checks your models, GPU and Blender |
workflows/PixelArtistry_01_Image_to_Watertight_Mesh.json |
Image → watertight 3D model (1536) – also print-ready |
workflows/PixelArtistry_01b_Image_to_Watertight_Mesh_2K.json |
Same at 2048 for maximum detail (16 GB+ VRAM, 32 GB+ RAM) |
workflows/PixelArtistry_02_Image_to_GameReady_Asset.json |
Image → textured high-poly + baked game-ready low-poly |
workflows/PixelArtistry_03_Your_Mesh_to_GameReady_Asset.json |
Your own mesh → textured + baked low-poly |
The installer already contains the workflows – you only need the .bat. The JSON files are here if you want to load them by hand.
- A working ComfyUI + TRELLIS.2 setup – follow the TRELLIS.2 Installation Guide
- Windows 10 or 11, NVIDIA RTX 20 series or newer
- ComfyUI Easy-Install with Python 3.12 and PyTorch 2.8.0 + CUDA 12.8
- VisualBruno's ComfyUI-Trellis2 installed – it provides CuMesh and O-Voxel
- Blender – ideally a separate, clean install without add-ons (see Troubleshooting)
- 32 GB system RAM recommended
- If you haven't yet, set up ComfyUI + TRELLIS.2 with the TRELLIS.2 Installation Guide.
- Download
watertightMeshes_win_installer.bat. - Put it into your
ComfyUI-Easy-Install\Add-onsfolder (the Easy-Install or portable root folder works too). - Close ComfyUI and double-click the file.
- When the summary shows only
[OK]lines, start ComfyUI and open Workflows → PixelArtistry. - In each workflow, paste your Blender path once into the purple Blender path node and load your image in Load Image.
What the installer does:
- Checks Python, PyTorch, CUDA and your GPU
- Downloads the node packs – pinned to the exact versions tested for the video, straight from the original repositories
- Picks the right WTiVo build for your GPU (RTX 50 or RTX 20/30/40)
- Installs the Python dependencies into
python_embeded - Runs a quick watertight test on your GPU
- Installs the four PixelArtistry workflows
- Checks all models – including shared model folders – and offers to download missing ones from Hugging Face
- Looks for Blender
It's safe to run again: everything is set back to the tested versions.
| VRAM | WTiVo resolution | Proxy points | Measured on an RTX 5080 |
|---|---|---|---|
| 6 GB | 1024 | lower if you run out of memory | – |
| 8 GB+ | 1536 (workflow 01/02) | 12,000,000 | WTiVo ~2 min, ~6 GB RAM |
| 16 GB+ | 2048 (workflow 01b, Quad Reconstruct bypassed) | 25,000,000 | WTiVo ~3.5 min, ~11 GB RAM |
At 2K, Quad Reconstruct is bypassed: its 2048 output made WTiVo's mesh balloon to 85M faces (~17 GB RAM with 25M proxy points), and LODTailor could no longer reduce it without breaking it. Without Quad, 25M proxy points give a 43M-face mesh that stays watertight through the whole chain.
- Blender errors in the console (Auto-Rig Pro, PolyQuilt, …): add-ons break headless Blender runs. Download the Blender zip, extract it, create an empty folder named
portablenext toblender.exe, and use that path in the Blender path node. - Holes after generation: run again – the structure seed is set to randomize.
- Checking watertightness in Blender: use the 3D Print Toolbox → Check All. Import textured low-poly files with Merge Vertices ticked – a GLB can't give one vertex two UV coordinates, so UV seams otherwise show up as open edges.
- 2K runs out of memory: restart ComfyUI and close other programs before the run.
- Installer says ComfyUI is still running: close it and run the installer again.
The watertight, low-poly and baking steps run on free, open-source nodes by MostAadTech – please support him: YouTube · Patreon · GitHub · X
| Node pack | Used for |
|---|---|
| WTiVo | making the mesh watertight |
| Mesh Quad Reconstruct | cleaning the raw mesh |
| LODTailor Mesh Trimmer | lighter and low-poly meshes |
| LODTailor Bake Forger | baking high-poly maps onto the low-poly |
| WTiVo Fast Merge by Distance | welding duplicate vertices |
| Trellis2 Mesh Encoder | reading your own mesh |
| CuMesh Decimate | GPU decimation |
| Memory Cleaner | freeing VRAM between steps |
| LODsmith Merge & Watertight | installed for completeness |
Also built on:
- ComfyUI's official Pixal3D / TRELLIS.2 template (MIT, © Comfy Org)
- VisualBruno's ComfyUI-Trellis2 (CuMesh, O-Voxel)
- Models from Comfy-Org/Pixal3D and Comfy-Org/TRELLIS.2 – originals by Tencent ARC and Microsoft
The installer and the workflows are MIT-licensed (see LICENSE). The workflows are based on ComfyUI's official template (MIT, © Comfy Org).
The installer doesn't redistribute any node pack or model – it downloads them from their original sources, and each keeps its own license (the node packs are GPL-3.0 or MIT). If you plan to use your results commercially, check the license of every node pack and model you use.
- Website: https://pixel-artistry.com
- YouTube: https://www.youtube.com/@PixelArtistry_
- X: https://x.com/philippsieben
- Newsletter FutureFrames: https://go.pixel-artistry.com/newsletter