Before you start
Separate upstream requirements from WSL, ComfyUI wrappers, and native community ports, then choose one path and keep its dependency chain isolated.
Prerequisites
- Windows 11 or a supported Windows build
- Compatible NVIDIA driver
- Administrator access for build tools or WSL
- Comfort reading issue trackers
Step by step
Choose WSL or a community-native path
WSL stays closer to Linux instructions; native wrappers may be easier to launch but use their own compiled dependencies.
Verify GPU visibility
Confirm the selected environment can see the GPU and expected CUDA runtime before installing model packages.
Pin the working stack
Record exact wrapper, PyTorch, CUDA, compiler, and custom-extension versions. Avoid upgrading one component in isolation.
Test the minimum pipeline
Run a supplied sample at the lowest practical setting and inspect the exported file before enabling optional post-processing.
Known trade-offs
- The official TRELLIS.2 repository currently documents Linux, not native Windows
- Community binaries should come from a project you can audit and verify
- A successful launch does not prove high-resolution workloads will fit in VRAM