MultiMatte is a promptable image background removal model developed by Feyn Labs that lets users specify what to keep using plain text prompts. Built on Meta's SAM 3 with LoRA fine-tuning, it produces alpha mattes instead of binary masks, enabling accurate handling of fuzzy edges like hair and translucent objects. MultiMatte significantly outperforms SAM 3 on standard benchmarks, scoring 0.901 S-measure on DIS-VD compared to SAM 3's 0.667. It is available as an open-source model on Hugging Face and installable via the NoBg Python library.