# On-device image AI — background remover, upscaler & more

> Remove backgrounds, upscale, blur faces, caption and depth-map images with AI that runs in your browser. Model weights download once; photos stay local.

Vision Studio runs neural image models in your browser. Weights download once and cache locally — the images you process never leave your device.

## Tools in this workspace

- [Background remover](https://www.safepaper.app/vision/background-remover): Cut out the subject and export a transparent PNG — the neural matting model runs on your device.
- [Image upscaler](https://www.safepaper.app/vision/image-upscaler): Super-resolve photos 2x with an on-device Swin2SR model. No upload, no watermark.
- [Low-light enhance](https://www.safepaper.app/vision/low-light-enhance): Brighten dark photos with local histogram stretching and gamma — pure canvas, no model download.
- [Object detector](https://www.safepaper.app/vision/object-detector): Run DETR object detection in your browser and see labelled bounding boxes over your image.
- [Auto blur](https://www.safepaper.app/vision/photo-blur): Detect people and objects, then blur the chosen classes before sharing — everything stays local.
- [Caption & alt-text](https://www.safepaper.app/vision/image-captioner): Generate a descriptive caption / alt-text for any image with an on-device vision-language model.
- [Zero-shot tagger](https://www.safepaper.app/vision/zero-shot-tagger): Ask 'does this image contain X?' with CLIP — score your own labels locally, no training needed.
- [Depth map](https://www.safepaper.app/vision/depth-map): Estimate a depth map from a single photo with an on-device DPT model.
- [Batch pipeline](https://www.safepaper.app/vision/batch-processor): Queue several images and apply background removal or upscaling to all of them on your device.

Each tool also has a Markdown mirror at its URL plus `.md` (for example https://www.safepaper.app/vision/background-remover.md).

---

Canonical HTML: https://www.safepaper.app/vision
Markdown: https://www.safepaper.app/vision.md

There is no upload endpoint — your files are processed in this browser tab. Open your network tab and check.