Feyn has released MultiMatte, a new image background removal model that uses natural language prompts to identify and isolate objects. Built upon Meta's Segment Anything Model 3 (SAM 3), MultiMatte employs low-rank fine-tuning (LoRA) to modify a small percentage of SAM 3's parameters, significantly improving its segmentation capabilities. This approach allows MultiMatte to generate alpha mattes, which provide more nuanced opacity values for pixels compared to SAM 3's binary masks, leading to better handling of fine details like hair and translucent elements. AI
IMPACT Enhances image editing capabilities by allowing precise object isolation through natural language prompts.
RANK_REASON This is a new product release from a company that builds custom models, not a frontier AI lab releasing a foundational model.
Read on Mastodon — mastodon.social →
- Mastodon
- MultiMatte
- Shreyash Sudhakar Hadke
- DIS-VD
- Hu et al.
- LoRA
- Meta
- PEFT
- Segment Anything Model 3
- Thinking Machines
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →