Skip to content

Wire Briefings Search

Fetching from the wire…

01

02

03

04

05

06

07

08

Wire Briefings Search

Entity trail

Feyn Labs

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

1

Findings

2

Edges

4

Sources

2

Corpus findings

2026-09-11 / projects-researcherMultiMatte turns SAM 3 into promptable background removal by fine-tuning 2.27 percent of its weightsFeyn Labs published MultiMatte on 2026-09-10. It keeps whatever a text phrase names ("the dog") and removes the rest, built by low-rank fine-tuning 19.49M of SAM 3's 860M parameters. It reports an S-measure of 0.901 on DIS-VD versus 0.667 for stock SAM 3, and 0.979 on the unseen DAVIS-S split. It ships as `pip install nobg`, with weights on Hugging Face. The benchmarks are the authors' own and the Show HN had 49 points.
2026-07-27 / hn-researcherShow HN: FeyNoBg Beats BiRefNet on 4 of 8 Background-Removal Benchmarks With 263M ParametersFeyn Labs released FeyNoBg on July 21 — a 263M-parameter matting model built by adding 6 blocks to the third feature-extractor stage of BiRefNet's 222M — alongside NoBg, the open-source Python training library behind it. On S-measure it leads on four benchmarks (UHRSD-TE 0.981 vs BiRefNet 0.957, HRSOD-TE 0.983 vs 0.961, DIS5K 0.961, DAVIS-S 0.977) and trails by under 2% on the other four (DUTS-TE, COD10K-TE, DUT-OMRON, CAMO-TE). It hit 90 points on HN; weights are on Hugging Face at feyninc/FeyNobg and the library at github.com/feyninc/nobg.

Graph relationships

BUILT BY
MultiMatte
Feyn Labs published MultiMatte by fine-tuning SAM 3.
Source finding
BUILT BY
FeyNoBg
FeyNoBg was developed by Feyn Labs.
Source finding
BUILT BY
MultiMatte -> Feyn Labs
Feyn Labs published MultiMatte by fine-tuning SAM 3.
Source finding
BUILT BY
FeyNoBg -> Feyn Labs
FeyNoBg was developed by Feyn Labs.
Source finding

Source trail

Feyn Labs released it July 21 Feyn Labs

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues