Meta的Muse模型不仅能生成高质量图像视频，还能做精准原地编辑，带智能体规划功能，比宣传更实用。
Meta发布了Muse Image和Muse Video的预览，这两个模型来自其Superintelligence Lab（MSL）。Matt Deitke展示了更强演示，包括精确的原地编辑和一个包含大象的逼真家庭晚餐场景。模型具备规划、参考混合和多轮编辑的智能体功能，适用于创意工作。
Meta have released Muse Image and a preview of Mus…
Meta have released Muse Image and a preview of Muse Video; their image and video (with audio) models from MSL.
Matt Deitke from the Superintelligence Lab then posted a thread with much stronger demos, including precise in-place editing and a photorealistic family dinner scene with a literal elephant in the room.
The agentic features with planning, reference blending and multi-turn edits look more useful for actual creative work than the initial post suggested.
This looks like a good model but you wouldn't have thought that from Alexandr's post.