Lance 是字节跳动开发的 3B 多模态模型，能够在单一架构内进行图像和视频理解、生成和编辑，并且完全从零开始训练。

一个 3B 的小模型，看图、看图说话、画图、看视频、生成视频全都能干。 Lance 是字节出的 3B 参数多模态模型，一个模型搞定图像和视频的理解、生成和编辑。除了 ViT 和 VAE 编码器，transformer 主干完全从零训练，128 张 A100 就能跑完。

Show HN: Lance – image/video generation and understanding in one model (

And here’s a reason why I need some serious GPU power in my home lab. Local inference for image and video work need me a some serious @NVIDIAAI action

WTF!! 3B model that can generate txt audio and video!!!! @BitDanceNetwork 🔥🔥🔥

Lance 🙂☺️by Bytedance - A 3B native (apache2.0 license) unified multimodal model for -image and video understanding -img + video generation, - and editing. page: Github- Hf repo-

ByteDance just dropped Lance. It is a 3B multimodal model for images and video that they actually trained from scratch. Staged multi-task training is the way to go if you want efficiency at this…