AI background blur
Blur the background while keeping the segmented subject sharp.
AI background blur
Drop your files here
or choose files from your device
How to use AI background blur
AI background blur keeps the segmented subject of a photo sharp while blurring everything behind it, similar to a portrait mode effect. The blur radius is adjustable from 1 to 100 pixels, with a default of 15. Segmentation uses the BEN2 model, downloaded on first run, and the result is saved as a PNG.
- Choose a source file. If the tool accepts multiple files, arrange them in the order you want.
- Set blur radius.
- Run ai background blur, review the output, then use the available copy or download controls.
What this tool supports
Downloads a browser inference model on first run, then processes source locally in a cancellable Web Worker. Background removal uses BEN2; transcription uses Whisper tiny. Review recognition errors and image edges.
- Blur radius
- Range: 1 to 100
Limits and processing
Image inputs: up to 32 megapixels, batches up to 10 images. Audio/video uploads: 50 MB; transcription: 120 seconds. Edited videos: 20 seconds, 4 fps and 640 pixels wide. Translation: English/Spanish, 30,000 characters; subtitles: 200 cues. Neural speech: 1,000 English characters. Model downloads and memory requirements vary by device.
Example source
A useful tool makes everyday work easier. Review the results before sharing.
Frequently asked questions
How do I blur the background of a photo but not the person?
Upload the photo and set the Blur radius. The model finds the main subject, the full image is blurred, and the sharp subject is drawn back on top, so only the background appears soft.
How strong should the background blur be?
The default of 15 pixels gives a moderate effect, and values toward 100 make the background unrecognizable. Higher values can also make segmentation mistakes more visible, so check the outline around hair and fine details.
Why is there a halo around my subject?
The blur is applied to the whole original image, including the subject, before the sharp cutout is placed on top. Where the mask misses part of the subject or has soft edges, blurred subject colors can show through as a halo.
Where is my input processed?
Source files and text stay in this browser. The first run downloads public models from Hugging Face and local runtime assets from the site. Models may remain in the browser cache. No source upload to Vercel, Supabase or a conversion service is required.
What are the input limits?
Image inputs: up to 32 megapixels, batches up to 10 images. Audio/video uploads: 50 MB; transcription: 120 seconds. Edited videos: 20 seconds, 4 fps and 640 pixels wide. Translation: English/Spanish, 30,000 characters; subtitles: 200 cues. Neural speech: 1,000 English characters. Model downloads and memory requirements vary by device.