Paper summary: Janus - Decoupling Visual Encoding for Unified Multimodal Understanding and Generation

Repo: https://github.com/deepseek-ai/Janus (MIT license!)

Paper: https://arxiv.org/html/2410.13848v1