Tiny-Engram: Trigger-Indexed Concept Tables for Generative Vision
The paper introduces Tiny-Engram, a method for enhancing generative vision models by using trigger-indexed concept tables. This approach allows for better control over the retrieval of visual memories within frozen image and video generators. The results indicate that this method can improve modular visual personalization, particularly in image generation, while highlighting the need for further development in video generation.
- ▪Tiny-Engram provides an explicit lexical address and activation boundary for visual memories in generative models.
- ▪The method uses small sets of memory entries indexed by n-gram matches to modulate text-encoder hidden states.
- ▪Evaluation shows that while image generation benefits significantly, video generation still requires improvements for identity persistence.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Computer Vision and Pattern Recognition arXiv:2605.20309 (cs) [Submitted on 19 May 2026] Title:Tiny-Engram: Trigger-Indexed Concept Tables for Generative Vision Authors:Runyuan Cai, Yiming Wang, Yu Lin, Xiaodong Zeng View a PDF of the paper titled Tiny-Engram: Trigger-Indexed Concept Tables for Generative Vision, by Runyuan Cai and Yiming Wang and Yu Lin and Xiaodong Zeng View PDF HTML (experimental) Abstract:Current personalization methods for generative vision models typically encode new concepts through continuous adapters or weight updates, yet provide limited control over whether and when a concept should be retrieved.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
