PropFly: Learning to Propagate via On-the-Fly Supervision from Pre-trained Video Diffusion Models
Wonyong Seo 1, Jaeho Moon 1, Jaehyup Lee 2, Soo Ye Kim 3, Munchurl Kim 1, 2 Kyungpook National University, 3 Adobe Research · 2026
Read PropFly: Learning to Propagate via On-the-Fly Supervision from Pre-trained Video Diffusion Models by Wonyong Seo 1, Jaeho Moon 1, Jaehyup Lee 2, Soo Ye Kim 3, Munchurl Kim 1, 2 Kyungpook National University, 3 Adobe Research (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-2602-20583.txt.
Description
Propagation-based video editing lets a user edit one frame and spread that edit through later frames, keeping motion and structure intact. Wonyong Seo and four co-authors note that training such models needs large paired video sets, costly and complex to gather. Their pipeline, PropFly, replaces those sets with on-the-fly supervision from pre-trained video diffusion models. It builds diverse latent pairs from one-step clean latent estimates at varied guidance scales.
The low-guidance latent carries the video's structure, while the high-guidance latent carries the target change. An added adapter on the pre-trained model then learns to spread edits through Guidance-Modulated Flow Matching loss, steering the model to copy the target change. Because supervision is made on the fly, the model learns time-consistent changes. Broad tests show PropFly clearly beats state-of-the-art methods on varied editing tasks.
Catalog record
| CREATOR | Wonyong Seo 1, Jaeho Moon 1, Jaehyup Lee 2, Soo Ye Kim 3, Munchurl Kim 1, 2 Kyungpook National University, 3 Adobe Research |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 60 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-20583