Turns the install/training README into a proper model card, without changing any of the original content:
- TL;DR and a Model description section (pixel back-projection conditioning, three-stage cascade, TRELLIS.2 backbone, DINOv2 encoder, GLB + PBR output, multi-view and scene extension)
- Intended use and Limitations/biases sections (generated-not-measured, unseen-surface invention, ObjaverseXL bias, view-aligned pose, VRAM modes)
- Quick start pointing to GitHub, with the one-image / low-VRAM / SDPA invocations
- Keeps the original Branches, Training, Community, Acknowledgements, Citation and License sections
Facts sourced from the paper (arXiv:2605.10922), the GitHub README, and the NOTICE file.
- Add image-to-3d pipeline tag and improve model card (45fafc48f156a48954cb4d3a2805e9df39c564a1)
Co-authored-by: Niels Rogge <nielsr@users.noreply.huggingface.co>
