01
Open an image and choose a model
Add a PNG, JPG, or WebP. Use the lightweight general model for an initial people or product cutout, or the portrait model for a person. Model assets download to your browser; the image stays local.
Remove image backgrounds with local AI for people and products
Try browser-local AI background removal for portraits and products without uploading your image. Download the model, inspect the cutout, and export a transparent PNG or a solid-background image.
Free to useNo watermarkFiles stay private
Keep the subject. Give it a new background.
A clear subject, without the background.
Drop a PNG, JPG, or WebP here, up to 20 MB and 16 megapixels. Try a clearly visible person, product, or object.
AI models download on first use: about 4.6 MB for general subjects or 25.9 MB for portraits, plus the local inference runtime.Models download to your browser. Your images stay on your device.
FROM SOURCE TO FINISHED FILE
How to use the AI background remover online
02
Download the model and remove the background
Watch model download and processing progress, and cancel if needed. Inference runs at the model working resolution; the resulting mask is applied back to the original image dimensions.
03
Inspect edges and export
Compare the original and result around hair, small gaps, and transparent objects. Download a transparent PNG, or choose a solid background and export PNG or JPG.
Make a subject cutout without a green screen
AI segmentation estimates a foreground subject from the image, so it can be useful for portraits and product photos with varied backgrounds. It does not need one selected key color. The lightweight general model is convenient to start with, while the portrait model is intended for people.
These are model estimates. Busy scenes, fine hair, low contrast, glass, multiple subjects, and shadows may produce missing details or leftover background. Review the original and cutout before using the result in a product listing or design.
Keep the image local while downloading model assets
The general U²-Netp model is served with the site and is approximately 4.6 MB. The optional MODNet portrait model is approximately 25.9 MB and downloads from a third-party model host. Downloads require a network connection; your chosen image is not sent to that host.
The model processes a working image at a lower resolution. Its mask is resized and combined with the source alpha at the original dimensions. Keeping output dimensions does not mean every fine edge was inferred at full resolution. Cached model files can be reused where browser storage permits.
What to know before using this AI background remover
- The lightweight model can miss small details or confuse complex backgrounds; portrait mode is intended for people and is not a universal product model.
- First use downloads model assets. The portrait model uses a third-party host; availability and browser caching can affect loading.
- Only still images are supported. Original dimensions are retained, but fine edges depend on the lower model working resolution; animation and metadata are not preserved.
PRACTICAL ANSWERS
AI background remover questions and answers
Formats, useful settings, and the details that affect your result.
Does my picture leave the device during AI background removal?+
No. Model files download into your browser, and inference runs locally. The portrait model host receives ordinary model download requests, but the selected image is not uploaded to it.
Which model should I use for a portrait or a product?+
Start with the lightweight general model for a product or a quick cutout. Try portrait mode for people. Compare results because neither model guarantees clean edges for every image.
Why does the first removal need a download?+
The browser needs the AI model and runtime assets before it can process your image. The tool shows progress and supports cancellation. Cached assets may make later runs faster where browser storage is available.
Is the exported image full size?+
The output keeps the accepted original width and height. The AI mask is inferred at a lower working resolution and resized, so fine detail is limited by the model rather than the output pixel count.
Can I replace the removed background with white or another color?+
Yes. Choose a solid background and export PNG or JPG. Choose the transparent option and PNG when you need a cutout for layering; JPG cannot store transparency.