MORPHOS: Autoregressive 4D Generation with Temporal Structured Latents
Minkyung Kwon, Jinhyeok Choi, Youngjin Shin, Jaeyeong Kim, JongMin Lee · 2026
Read MORPHOS: Autoregressive 4D Generation with Temporal Structured Latents by Minkyung Kwon, Jinhyeok Choi, Youngjin Shin, Jaeyeong Kim, JongMin Lee (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-02491.txt.
Description
Kwon and co-authors present MORPHOS, a system that generates moving 3D assets from video. Older methods bind themselves to one representation. They struggle with topology changes. They also lose temporal consistency over long clips. MORPHOS targets all three gaps at once.
The core is a unified 4D representation called Temporal Structured Latents. It encodes geometry and appearance along the time axis. MORPHOS generates assets step by step with causal attention. Each frame is conditioned on the history before it. Outputs can be meshes, 3D Gaussians, or radiance fields. A temporal-structural augmentation limits error buildup across frames. Tests show top appearance scores and strong geometry scores. Generalization holds across representations. Robustness holds over long horizons. The work offers one frame for dynamic 3D creation across output types.
Catalog record
| CREATOR | Minkyung Kwon, Jinhyeok Choi, Youngjin Shin, Jaeyeong Kim, JongMin Lee |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 58 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.02491