TokenDial: Continuous Attribute Control in Text-to-Video via Spatiotemporal Token Offsets
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866911550885855232 |
|---|---|
| author | Liu, Zhixuan Schaldenbrand, Peter Li, Yijun Mai, Long Mahapatra, Aniruddha Ham, Cusuh Oh, Jean Wang, Jui-Hsien |
| author_facet | Liu, Zhixuan Schaldenbrand, Peter Li, Yijun Mai, Long Mahapatra, Aniruddha Ham, Cusuh Oh, Jean Wang, Jui-Hsien |
| contents | We present TokenDial, a framework for continuous, slider-style attribute control in pretrained text-to-video generation models. While modern generators produce strong holistic videos, they offer limited control over how much an attribute changes (e.g., effect intensity or motion magnitude) without drifting identity, background, or temporal coherence. TokenDial is built on the observation: additive offsets in the intermediate spatiotemporal visual patch-token space form a semantic control direction, where adjusting the offset magnitude yields coherent, predictable edits for both appearance and motion dynamics. We learn attribute-specific token offsets without retraining the backbone, using pretrained understanding signals: semantic direction matching for appearance and motion-magnitude scaling for motion. We demonstrate TokenDial's effectiveness on diverse attributes and prompts, achieving stronger controllability and higher-quality edits than state-of-the-art baselines, supported by extensive quantitative evaluation and human studies. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_27520 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | TokenDial: Continuous Attribute Control in Text-to-Video via Spatiotemporal Token Offsets Liu, Zhixuan Schaldenbrand, Peter Li, Yijun Mai, Long Mahapatra, Aniruddha Ham, Cusuh Oh, Jean Wang, Jui-Hsien Computer Vision and Pattern Recognition We present TokenDial, a framework for continuous, slider-style attribute control in pretrained text-to-video generation models. While modern generators produce strong holistic videos, they offer limited control over how much an attribute changes (e.g., effect intensity or motion magnitude) without drifting identity, background, or temporal coherence. TokenDial is built on the observation: additive offsets in the intermediate spatiotemporal visual patch-token space form a semantic control direction, where adjusting the offset magnitude yields coherent, predictable edits for both appearance and motion dynamics. We learn attribute-specific token offsets without retraining the backbone, using pretrained understanding signals: semantic direction matching for appearance and motion-magnitude scaling for motion. We demonstrate TokenDial's effectiveness on diverse attributes and prompts, achieving stronger controllability and higher-quality edits than state-of-the-art baselines, supported by extensive quantitative evaluation and human studies. |
| title | TokenDial: Continuous Attribute Control in Text-to-Video via Spatiotemporal Token Offsets |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2603.27520 |