LiteFrame: Efficient Vision Encoders Unlock Frame Scaling in Video LLMs

Jihwan Kim^1,2, Nikhil Parthasarathy¹, Danfeng Qin¹, Junhwa Hur¹, Deqing Sun¹, Bohyung Han^1,2, Ming-Hsuan Yang¹, Boqing Gong¹

¹Google DeepMind        ²Seoul National University

          

   

---

TL;DR: We propose LiteFrame, a highly efficient video encoder for Video Large Language Models that unlocks scalable, long-form video understanding by resolving inefficiencies in both the LLM and the ViT.

🚧 Note: Code and model weights will be released soon.

1-Min Overview 🚀

LiteFrame.mp4

News 📰

• [2026.05.18] Our paper, LiteFrame: Efficient Vision Encoders Unlock Frame Scaling in Video LLMs, has been archived.

Citation 📃

If you find our work useful for your research, please consider citing:

@article{kim2026liteframe,
  title={LiteFrame: Efficient Vision Encoders Unlock Frame Scaling in Video LLMs},
  author={Kim, Jihwan and Parthasarathy, Nikhil and Qin, Danfeng and Hur, Junhwa and Sun, Deqing and Han, Bohyung and Yang, Ming-Hsuan and Gong, Boqing},
  journal={arXiv preprint arXiv:2605.17260},
  year={2026}
}