AI Image Tools
Remove backgrounds and extract text from images with AI — right in your browser.
Two jobs on this site need a model rather than a filter: separating a subject from its background, and reading text out of a picture. Both normally mean uploading your image to somebody's server. Here they do not — the models run inside your browser tab, downloaded once and then executed on your own device, so the photo you are working on never leaves it.
What each tool does
- Background Remover
- You need a subject on a transparent background — a product shot, a profile picture, a cut-out for a slide or a poster.
- Image to Text
- You have text trapped in a screenshot, a photo of a page, a receipt or a slide, and you want to copy and paste it.
How background removal actually works
The background remover runs U²-Net, an open-source salient-object detection model released under the Apache 2.0 licence. It does not recognise what your subject is; it predicts, pixel by pixel, how likely each one is to belong to the thing the picture is about, and that prediction becomes the transparency mask. This is why it handles a person, a shoe and a houseplant equally well, and why it struggles with the same things every model of its kind struggles with: fine hair, motion blur, glass, and subjects that are the same colour as what is behind them. A subject that stands out clearly from its background gives a clean edge almost every time.
How the text extraction works
Image to Text runs Tesseract, the long-established open-source OCR engine, compiled to WebAssembly so it executes in the browser. It works line by line on the shapes it finds, which makes it strong on clean, high-contrast, roughly horizontal text — screenshots, scanned documents, printed pages — and weak on handwriting, heavily stylised fonts, low-resolution photos and text at an angle. Getting a good result is mostly about the input: a sharp, straight, well-lit image of the text will beat a blurry one no matter how the engine is tuned.
The trade-off of running models locally
The first time you open one of these tools it downloads the model — a few megabytes for background removal — and that download is the slow part. After that it is cached and subsequent images are quick. Processing then happens on your CPU rather than a server GPU, so a large image takes seconds rather than being instant. What you get in return is complete: no upload, no account, no per-image quota, no retention policy to read, and no question about whether your photo becomes somebody's training data.
Getting a better result
For background removal, use the highest-resolution version you have and prefer an image where the subject is clearly separated from what is behind it — the model has more to work with than you might expect, but it cannot invent an edge that is not visible. For OCR, crop to just the text before running it, make sure the lines are horizontal rather than skewed, and avoid photographing a screen where possible, since moiré patterns confuse the engine badly. If a page is dim, run it through Brightness & Contrast first.
Frequently asked questions
Do these AI tools upload my image?
No. Both models are downloaded to your browser and run there. The image is processed on your own device and never sent anywhere, which is the main reason these tools exist in this form.
Why is the first run slow?
The model file has to download the first time — a few megabytes. It is then cached by your browser, so every image after that starts immediately.
What format should I download a cut-out subject as?
PNG or WebP. Both keep transparency. Saving as JPG fills the transparent area with a solid colour, which defeats the purpose.
Can OCR read handwriting?
Not reliably. Tesseract is built for printed text. Neat block capitals sometimes come through, but cursive and casual handwriting generally do not.
Does it work offline?
Once the model is cached, yes — with the tab already open, both tools keep working without a connection.