Skip to content
← AI Tools

AI Smart Cutout

Click any object in your image to lift it out as a clean transparent PNG. Object-level AI selection powered by Segment Anything (SlimSAM), running entirely in your browser — no upload, no signup, no API key.

AI Smart Cutout

Click any object in an image and lift it out as a clean transparent PNG. Smart Cutout uses Segment Anything to turn a single click into a precise selection, so you can isolate exactly the thing you want — not just the whole foreground — without leaving your browser.

Why Object-Level Selection Beats One-Click Background Removers

A traditional background remover makes one automatic call: keep the foreground, drop the rest. That is perfect when there is a single clear subject, and our background remover handles that case well. But a lot of real images are not that tidy — three people standing together, one product on a crowded shelf, a sign you want to pull off a wall. Automatic tools either grab all of it or merge the thing you want into the background. Smart Cutout is interactive instead. You click the specific object, the model proposes a mask, and you refine it with a second click or by switching to Remove mode. This is the same capability behind paid features like Photoshop Select Subject and the object tools in remove.bg or Photoroom — offered here free, with nothing uploaded.

How AI Smart Cutout Works

Click Load model on your first visit. The browser downloads the quantized SlimSAM weights (about 40 MB) from the Hugging Face CDN and caches them in IndexedDB. Drop or pick an image, and the tool encodes it once — this is the only slow step, because the vision encoder runs a single time per image. After that, every click is near-instant: each click runs only the lightweight mask decoder, which proposes a segmentation mask for the point you clicked. Add more points to extend the selection, or right-click (or switch to Remove mode) to carve away anything the model over-selected. When the highlighted mask matches what you want, press Cut out selection to bake the mask into the alpha channel and export a transparent PNG. The original image is untouched, so you can re-select a different object, then send the cutout straight into a crop or composite. For depth-based masking instead of object-based, the depth estimator is the companion tool. To delete an object and rebuild the background behind it instead of extracting it, use the AI Magic Eraser.

The Model, the License, and Your Privacy

The tool runs SlimSAM, a compressed build of Meta AI Segment Anything Model (SAM). The original SAM is large and impractical to run fully client-side; SlimSAM shrinks it dramatically while keeping strong segmentation quality, which is what makes in-browser inference feasible. SAM is released under the Apache 2.0 license, which permits commercial use — the reason it can ship here free on an ad-supported site, where non-commercial or research-only model licenses are off limits. The ONNX weights are mirrored by the Xenova organization on Hugging Face and run through transformers.js, with WebGPU used when your browser supports it and a WebAssembly fallback otherwise. Because everything happens locally, your image is never uploaded — not to our servers, not to the model host, not to any API.

Frequently Asked Questions

How is Smart Cutout different from a background remover?+
A background remover makes one automatic decision: keep the whole foreground, drop everything else. Smart Cutout is interactive and object-level, so you click the specific thing you want. That lets you pull one person out of a group, lift a single product off a busy shelf, or isolate an object the automatic tools merge into the background. Add more clicks to extend the selection, or switch to Remove mode to subtract anything the model included by mistake.
How big is the download and is it cached?+
The model used here (SlimSAM, a compressed Segment Anything) is a compact download of roughly 40 MB, split across a vision encoder and a lightweight mask decoder. It downloads once from the Hugging Face CDN and is cached in your browser through IndexedDB, so every later visit loads it in a second or two.
Are my images uploaded to a server?+
No. Once the model has downloaded on first use, both the image encoding and every mask you generate run entirely in your browser. The image never leaves your device and is never sent to our servers or to any third-party API. You can disconnect from the internet after the model loads and the tool keeps working.
Which model powers this and what is its license?+
It uses SlimSAM, a compressed version of Meta AI Segment Anything Model (SAM), with ONNX weights mirrored by the Xenova organization on Hugging Face. SAM is released under the Apache 2.0 license, which permits commercial use. That clean license is the reason this can be offered free on an ad-supported site.
How do I select exactly the object I want?+
Click once on the object to get an initial selection. If the model grabs too little, click another spot on the same object to extend it. If it grabs too much, switch the click mode to Remove (or right-click on desktop) and click the area you want excluded. Each click refines the mask in place, and Clear points starts over.
Why is the first click on a new image slow?+
The heavy step is encoding the image, which the vision encoder does once per image right after you load it. After that, each click only runs the small mask decoder, which is near-instant. So the first selection waits on encoding, and every refinement afterward is fast.
What image formats and sizes work?+
PNG, JPEG, WebP, and GIF (first frame) all work. There is no server limit since processing is local, but very large images take longer to encode and use more memory, especially on the WebAssembly fallback path. Browsers with WebGPU encode noticeably faster.
Can I use the cutouts commercially?+
Yes. The model license (Apache 2.0) permits commercial use, and the cutout you export is just your own image with a transparent background, so its rights are whatever they already were for your original image. Nothing about this tool adds a watermark or a usage restriction.

Built by Derek Giordano · Part of Ultimate Design Tools

Privacy Policy · Terms of Service