AI background removal
Create a transparent PNG or replace the background with a clean studio color.
01PRIVATE AI PHOTO STUDIO
Studio-quality portraits.
Only in your browser.
Remove a background, create ID or passport photos, and brighten only the person—with no upload, account, watermark or server processing.
Photos stay on this device · models load only when used
Local portrait pipelineLOCAL
Your photo
1
U²-NetAI cutout
2
Final photo
Your photoNo photo upload
✓The page downloads an AI model, but your selected image is never uploaded. Preview and download are created in device memory.
01No server upload
02WebGPU + WASM fallback
03Free · no watermark
04PNG · JPG download
PHOTO WORKSPACE
One photo, four ways to finish
AI isolates the foreground once. Then you can remove, replace, size or retouch without sending the photo anywhere.
Your photo is processed only in this browser
LOCAL ONLYThe page downloads an AI model, but your selected image is never uploaded. Preview and download are created in device memory.
+
Drop your portrait here
PNG, JPG or WebP · up to 24 megapixels recommended
You can also paste an image with Ctrl + Vphoto.jpg
Choose a result
Live preview—
Model has not been loadedChoose a photo to start
iThe first AI run downloads and caches the selected model in this tab.
Downloading AI model…0%
ONNX Runtime Web. WebGPU is used when available; WebAssembly provides the fallback. Advanced U²-Net is split into static files below Cloudflare Pages’ 25 MiB per-asset limit.
LOCAL-FIRST
Portrait work without a photo server
The familiar Kivo workflow, now with local AI segmentation and live compositing.
ID photo canvas
Fit the isolated person into a 3 × 4 cm, 300 dpi output canvas.
02Passport photo canvas
Export a 3.5 × 4.5 cm, 300 dpi image with an on-screen face guide.
03Portrait-only beauty
Brightness and softening are composited through the person mask, leaving the background untouched.
04WebGPU with WASM fallback
Uses available GPU acceleration and automatically falls back to CPU-compatible WebAssembly.
05Private by architecture
There is no upload endpoint. Image pixels remain in the current browser tab.
0601 — 03
Three local steps
- 01
Choose a portrait
The browser reads the file directly from your device.
- 02
Run the AI cutout
ONNX Runtime Web executes U²-NetP or U²-Net locally.
- 03
Adjust and download
Canvas creates the final PNG or JPG in memory.
PRIVACY BY ARCHITECTURE
The model comes to the photo. The photo never goes to a server.
PhotoKivo is a static website. Only public app files and the selected AI model are downloaded; your photo and the generated mask remain local.
HELP
Frequently asked questions
Is my photo uploaded?
No. The file input, ONNX inference, preview and export run locally. Network requests are used only for static site files and AI model parts.
Why is the advanced model slow the first time?
The full U²-Net model is about 176 MB and must be downloaded before inference. The basic U²-NetP model is much smaller and is recommended for most edits.
Are passport photos guaranteed to be accepted?
No. The preset provides size and composition guides, but requirements differ by country, authority, age and application method. Check the official rules before submitting.
Which browsers work best?
Current Chrome or Edge offers the best chance of WebGPU acceleration. Other modern browsers can use the WebAssembly fallback.