Home

Lance: Unified Multimodal Modeling by Multi-Task Synergy

Fengyi Fu^*, Mengqi Huang^*†, Shaojin Wu^*, Yunsheng Jiang^*, Yufei Huo, Hao Li, Yinghang Song, Fei Ding, Jianzhu Guo^†§, Qian He, Zheren Fu, Zhendong Mao, Yongdong Zhang
^* Equal contribution, ^† Corresponding authors, ^§ Project lead

arxiv / paper / huggingface / homepage / code /

We present Lance, a 3B native unified multimodal model for image and video understanding, generation, and editing. Lance is trained from scratch with a staged multi-task recipe, combining unified context modeling and decoupled capability pathways. The code and model weights are available at https://github.com/bytedance/Lance and Hugging Face.