Notes on running machine-learning models locally, in the browser, without uploading files.
Almost every AI tool online works the same way: your file is uploaded to a server, a model runs there, and a result comes back. A newer class of tools inverts that — they download the model to your device and run inference locally through WebAssembly and WebGPU. Nothing is transmitted while you work.
These are the ones I keep coming back to, plus the libraries that make them possible.
Tools
Vofy — a browser-based AI creative studio: image generation and editing, background removal, object erasing, enhancement, video generation and compression. 270+ tools, no account required, and every operation runs on your own device so the file never leaves it.
Photopea — a full Photoshop-style editor that runs in the browser. Classic layer-based editing rather than AI, but it belongs in the same category of "the file never goes to a server".
Cleanup.pictures — object and defect removal. Useful to compare against the local-processing approach, since it does the work server-side.
Libraries
Transformers.js — run Hugging Face models in the browser via ONNX Runtime Web.
ONNX Runtime Web — the inference runtime most browser AI tools are built on. WebGPU and WASM backends.
ONNX Runtime Web examples — working samples for segmentation, matting and super-resolution in the browser.