PART OF RUNTIMEHUB · FREE · PRIVATE · INSTANT
Portrait isolation is a critical step in dozens of workflows — product listings, social profiles, video production, HR headshots, course thumbnails. Until now, doing it properly meant paying for cloud software that uploads your images to a server you don't control, or installing desktop apps that cost hundreds per year.
SoloBG was built to fix that. A specialized AI model trained exclusively on human subjects, running directly on your device, producing full-resolution transparent PNGs with no server, no account, and no cost.
The result is a tool that respects your data by design — not by policy. There is no server to breach, no database to subpoena, no third party to trust. Your images never leave your machine because they physically cannot.
At the core of SoloBG is MODNet, a neural network trained specifically on human portraits. Unlike general-purpose background removal models, MODNet understands human features at a structural level — hair strands, glasses, semi-transparent fabrics, motion blur at the edges.
The model runs via Transformers.js inside a Web Worker, keeping inference off the main thread. On first visit the model (~12 MB, quantized q4f16) downloads from Hugging Face and is cached locally in IndexedDB — subsequent visits load instantly with no network request.
Output is composited at the full input resolution — a 4K input produces a 4K transparent PNG. The in-browser preview is a scaled-down representation for performance; the download is always full resolution.
SoloBG is one tool in a growing suite of browser-native utilities built under RuntimeHub. Every tool in the suite shares the same principles: no server processing, no account required, no watermarks, full output resolution.
Other tools in the ecosystem include ZeroBG (general background removal), PixelBoost (4× AI upscaling), PixelTruth (AI image detection), and a dozen more. Each is free, private, and runs entirely in your browser.