# LLaVA-OneVision: Easy Visual Task Transfer

**Authors:** Bo Li, Yuanhan Zhang, Dong Guo, Renrui Zhang, Feng Li, Hao Zhang, Kaichen Zhang, Yanwei Li, Ziwei Liu, Chunyuan Li  
**arXiv:** [2408.03326](https://arxiv.org/abs/2408.03326)  
**Date:** 2024-08-06  
**Summary page:** https://www.summarizepaper.com/en/paper/2408.03326v1/

## Key points

- LLaVA-OneVision introduced as a family of open large multimodal models (LMMs) for computer vision
- Developed by a team including Bo Li, Yuanhan Zhang, Dong Guo, and others
- Demonstrated groundbreaking capabilities in single-image analysis, multi-image processing, and video understanding
- Robust transfer learning abilities across modalities and scenarios
- Impressive prowess in video comprehension and seamless task transfer from image-based to video-based contexts
- Sets a new standard for multimodal model development

## Summary

The paper "LLaVA-OneVision: Easy Visual Task Transfer" introduces LLaVA-OneVision as a family of open large multimodal models (LMMs) designed to push the performance boundaries in computer vision. Developed by Bo Li, Yuanhan Zhang, Dong Guo, Renrui Zhang, Feng Li, Hao Zhang, Kaichen Zhang, Yanwei Li, Ziwei Liu and Chunyuan Li through insights gathered from data and visual representations in the LLaVA-NeXT blog series. The experimental results demonstrate the model's groundbreaking capabilities in single-image analysis, multi-image processing and video understanding. Its robust transfer learning abilities across modalities and scenarios allow for new functionalities to emerge. Particularly impressive is its prowess in video comprehension and seamless task transfer from image-based to video-based contexts. This innovative approach sets a new standard for multimodal model development and can be accessed through their project homepage at https://llava-vl.github.io/blog/2024-08-05-llava-onevision/.