Consistent Video-to-Video Translation via Explicit Correspondences
arXiv cs.AI · 13 Nov 2026, 04:55 AEST
A new approach called vid2vid-long enhances video-to-video translation by ensuring long-range temporal consistency. This method utilizes patch-level correspondences to improve the quality of generated outputs in real-time applications.
Key facts
- Interactive video-to-video applications require real-time generation while maintaining long-range temporal consistency.
- To address this limitation, we introduce vid2vid-long, a correspondence-based approach grounded in a simple observation.
- Concretely, we compute patch-level correspondences in the input, guiding retrieval of previously generated outputs in the corresponding regions.
- Our experiments demonstrate that vid2vid-long consistently improves long-range temporal consistency across recent video models while preserving generation quality and real-time performance.
Read aloud
A new approach called vid2vid-long enhances video-to-video translation by ensuring long-range temporal consistency. This method utilizes patch-level correspondences to improve the quality of generated outputs in real-time applications. Interactive video-to-video applications require real-time generation while maintaining long-range temporal consistency. To address this limitation, we introduce vid2vid-long, a correspondence-based approach grounded in a simple observation. Concretely, we compute patch-level correspondences in the input, guiding retrieval of previously generated outputs in the corresponding regions. Our experiments demonstrate that vid2vid-long consistently improves long-range temporal consistency across recent video models while preserving generation quality and real-time performance.
Sources in this cluster
- Consistent Video-to-Video Translation via Explicit Correspondences
arXiv cs.AI · 13 Nov 2026, 04:55 AEST