July 27, 2026
Cutout chaos, but make it open-source
Show HN: FeyNoBg – Automatic background removal model and training library
Open-source photo cutout tool drops, and commenters are already asking: can it go bigger and go mobile?
TLDR: FeyNoBg is a new open-source tool that removes image backgrounds and claims top-tier results, with its full training toolkit released too. Commenters loved the simplicity but instantly pushed for answers on bigger images, mobile use, and whether it can replace fussier alternatives.
A new open-source photo cutout tool called FeyNoBg just landed, promising cleaner background removal so your subject stays sharp while the messy stuff behind them disappears. The makers say it performs at the top of the pack across major tests, and they also released the training library so other people can build their own versions too. In plain English: this isn’t just a neat app, it’s the recipe book as well.
But the real show is in the comments, where the vibe is a mix of hype, relief, and instant feature requests. One user basically said, finally, a simple tool they can actually use, taking a swipe at how another popular image-cutting tool has become annoying to work with. That sparked the thread’s main mood: people are less interested in benchmark bragging and more interested in whether this thing saves them time right now. Another commenter jumped straight into detective mode, asking why the readme mentions one image size limit when a much bigger image seemed to work fine anyway. Translation: the crowd immediately started stress-testing the claims.
Then came the biggest power-user energy of all: “Can this run on phones, and can I try porting it?” That turned the launch into a mini casting call for hackers who want this everywhere. There wasn’t much outright fighting, but there was classic Hacker News subtext: congratulations first, then practical grilling second. The funniest part is how fast the conversation moved from “nice release” to “cool, but will it handle giant files and fit in my pocket?”
Key Points
- •The article introduces FeyNoBg, an open-source automatic background removal model, and NoBg, the open-source library used to train it.
- •FeyNoBg is reported to achieve the best published S-measure on four of eight benchmarks and to come within 2% of the leader on the other four.
- •The article explains background removal as requiring both foreground recognition and accurate boundary estimation, including image matting for partial-opacity edge pixels.
- •FeyNoBg is built on BiRefNet, with the third stage of the feature extractor expanded from 18 to 24 blocks, increasing parameter count from 222M to 263M while preserving compatible pre-trained weights.
- •An initial training run using only the synthetic MaskFactory dataset improved CAMO results but regressed on DIS5K, leading the authors to assemble a more diverse dataset of 26.1K images from 10 datasets.