Feyn has introduced FeyNoBg, a state-of-the-art background removal model, alongside NoBg, an open-source Python library designed to streamline the training and inference of such models. Built upon the BiRefNet architecture, FeyNoBg utilizes a localization module to identify subjects and a reconstruction module to trace boundaries. To enhance boundary precision, the developers expanded the third stage of the feature extractor from 18 to 24 blocks, increasing the model's parameter count from 222M to 263M. The training strategy focused on overcoming the imbalance between foreground recognition and edge precision by using a diverse dataset of 26.1K images curated from 10 sources, including S3OD and various portrait and anime datasets. To prevent larger datasets from dominating, each source was capped at 4,000 images. All annotations, including alpha mattes, were converted into binary foreground masks to ensure consistent training targets. In evaluations using the S-measure metric, FeyNoBg achieved the highest published scores on four out of eight benchmarks—covering scenarios like camouflage and high resolution—and remained within 2% of the leaders on the remaining four. Additionally, the NoBg library demonstrates superior performance over original BiRefNet implementations, offering higher throughput, lower latency, and reduced peak GPU memory usage during inference.