TRELLIS.2 runs fully in-browser on WebGPU: text to 3D in about 3 minutes
A developer ported the TRELLIS.2 text-to-3D pipeline entirely into the browser using WebGPU via the C# library SpawnDev.ILGPU.ML 5.3.12. On an RTX 4070 in Chrome, generating a GLB takes 134–180 seconds, with ~10 GB of weights cached in OPFS and streamed via range requests.
- Photo-to-GLB takes 148 s cold and 134 s warm on an RTX 4070
- Pipeline uses SD Turbo, RMBG-1.4 and Microsoft's MIT-licensed TRELLIS.2-4B
- Weights cached in OPFS with four 16 MB range requests in flight
- AOT cut mesh simplification from 41.3 s to 2.6 s
Read next
AI