# CRAFT: Compression via Recursive Adaptive Fusion of Video Tokens for Vision-Language Models

arXiv: https://arxiv.org/abs/2608.01644 · HTML: https://mlsanity.com/paper/2608.01644/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Visual token pruning and efficiency, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 23 of its 38 references that are in the map.
- First posted on arXiv 3 Aug 2026.

## Placed in

- Multimodal LLMs / vision-language models / [Visual token pruning and efficiency](https://mlsanity.com/mllm/visual-token-pruning-and-efficiency/)

## Builds on (most cited first, 23 of 23)

- [Masked Autoencoders Are Scalable Vision Learners](https://mlsanity.com/paper/2111.06377/) (2021)
- [Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution](https://mlsanity.com/paper/2409.12191/) (2024)
- [LLaVA-OneVision: Easy Visual Task Transfer](https://mlsanity.com/paper/2408.03326/) (2024)
- [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://mlsanity.com/paper/2203.12602/) (2022)
- [MVBench: A Comprehensive Multi-modal Video Understanding Benchmark](https://mlsanity.com/paper/2311.17005/) (2023)
- [Token Merging: Your ViT But Faster](https://mlsanity.com/paper/2210.09461/) (2022)
- [DynamicViT: Efficient Vision Transformers with Dynamic Token Sparsification](https://mlsanity.com/paper/2106.02034/) (2021)
- [EgoSchema: A Diagnostic Benchmark for Very Long-form Video Language Understanding](https://mlsanity.com/paper/2308.09126/) (2023)
- [LongVideoBench: A Benchmark for Long-context Interleaved Video-Language Understanding](https://mlsanity.com/paper/2407.15754/) (2024)
- [Not All Patches are What You Need: Expediting Vision Transformers via Token Reorganizations](https://mlsanity.com/paper/2202.07800/) (2022)
- [TempCompass: Do Video LLMs Really Understand Videos?](https://mlsanity.com/paper/2403.00476/) (2024)
- [LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding](https://mlsanity.com/paper/2410.17434/) (2024)
- [LongVILA: Scaling Long-Context Visual Language Models for Long Videos](https://mlsanity.com/paper/2408.10188/) (2024)
- [Video-XL: Extra-Long Vision Language Model for Hour-Scale Video Understanding](https://mlsanity.com/paper/2409.14485/) (2024)
- [TokenPacker: Efficient Visual Projector for Multimodal LLM](https://mlsanity.com/paper/2407.02392/) (2024)
- [TimeChat-Online: 80% Visual Tokens are Naturally Redundant in Streaming Videos](https://mlsanity.com/paper/2504.17343/) (2025)
- [Token Reduction Should Go Beyond Efficiency in Generative Models -- From Vision, Language to Multimodality](https://mlsanity.com/paper/2505.18227/) (2025)
- [Multi-Granular Spatio-Temporal Token Merging for Training-Free Acceleration of Video LLMs](https://mlsanity.com/paper/2507.07990/) (2025)
- [LLaVA-Scissor: Token Compression with Semantic Connected Components for Video LLMs](https://mlsanity.com/paper/2506.21862/) (2025)
- [InternVL-X: Advancing and Accelerating InternVL Series with Efficient Visual Token Compression](https://mlsanity.com/paper/2503.21307/) (2025)
- [Less Is More, but Where? Dynamic Token Compression via LLM-Guided Keyframe Prior](https://mlsanity.com/paper/2512.06866/) (2025)
- [DynTok: Dynamic Compression of Visual Tokens for Efficient and Effective Video Understanding](https://mlsanity.com/paper/2506.03990/) (2025)
- [MARC: Memory-Augmented RL Token Compression for Efficient Video Understanding](https://mlsanity.com/paper/2510.07915/) (2025)

## Abstract (arXiv)

In video understanding, vision-language models (VLMs) must ingest massive numbers of visual tokens, causing the computational and memory cost of the prefill stage to rise sharply. Such visual sequences are highly redundant along the spatio-temporal dimension, yet a high compression ratio is often accompanied by the loss of critical details. Existing token-compression methods either employ heuristic, training-free compression with limited content adaptivity or introduce additional modules that require expensive alignment training, leaving the trade-off between efficiency and adaptivity unresolved. To alleviate this limitation, we propose CRAFT: Compression via Recursive Adaptive Fusion of Video Tokens. CRAFT recursively merges tokens by decoupling parameter-free token selection from learnable token fusion: global similarity determines which tokens to merge, while a position-aware weighting module and a content-adaptive channel-wise gate learn how to fuse them. The whole compression pipeline is query-agnostic. Because every retained token is a linear combination of the original tokens, CRAFT preserves their true spatio-temporal coordinates and stays aligned with the pre-trained language model's input distribution. Experiments on multiple representative video benchmarks show that CRAFT consistently outperforms prior state-of-the-art token-compression methods. At about 8× compression, it retains roughly 97% of the backbone's average accuracy and shows significant efficiency improvement.