MATE4D: Matrix-Guided Editable 4D Generation from a Single Image
Generative models have rapidly pushed content creation be-yond 2D imagery toward dynamic 3D and 4D scene synthesis. Yet pro-ducing realistic and temporally stable 4D content from a single image is still difficult because one view provides limited structural cues and weak motion evidence. We introduce MATE4D, a framework that converts one input image into editable dynamic 4D content. Our method constructs a spatio-temporal multi-view image matrix with text-guided background manipulation, delivering coherent supervision over viewpoint, appear-ance, and motion. These synthesized observations are used to optimize 3D Gaussian primitives, which are then animated through a lightweight deformatio...
| Authors | Xiaotian Chen, Dongfu Yin |
|---|---|
| Primary category | cs.CV |
| Published | 8 October 2026 |
| Last updated at source | 8 October 2026 |
Popularity, last 2 days
0 now. Counts as reported by the source, snapshotted daily by Silytics.
Silytics shows public metadata from arXiv and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.