The browser checks the source.
FlytheBG validates the selected file and confirms that the browser can decode it. PNG, JPEG, and WebP can go straight through; other raster formats can be normalized locally when the browser supports them.
Remove Background
Keep the frame. Remove the background.
Use portrait, landscape, square, vertical, or panoramic raster images without forcing a preset crop. FlytheBG preserves the source aspect ratio, uses the smaller IMG.LY quantized model, and tries WebGPU first where supported.
Any normal ratiosource proportions preservedGPU → CPUautomatic compute fallbackPNGtransparent output
Browser AI
● No image uploadDrop an image. Keep it on your device.
Most browser-decodable raster image formats up to 12 MB.
01IMG.LY QuantizedWebGPU when available · CPU fallback
Runs on the visitor's device.FlytheBG serves the static app; the browser downloads the small IMG.LY model/runtime assets only when needed.
Background removal guide
What happens after you choose an image.
Background removal is foreground segmentation, not a simple color delete. The model estimates an alpha mask describing how visible each foreground pixel should remain, and FlytheBG returns the resulting transparent PNG without changing the source aspect ratio.
The small model runs with a compute fallback.
The current implementation uses IMG.LY’s quantized IS-Net model. Compatible secure browsers try WebGPU first for acceleration. If that backend cannot initialize or finish, FlytheBG retries the same small model using CPU/WASM.
The model cutout is validated and exported.
FlytheBG checks that the generated result contains meaningful foreground and transparency, then uses the model cutout directly. The old full-resolution edge-expansion pass is no longer part of the live workflow.
Better source photos
How to get cleaner edges before you upload.
The model can only work with detail that exists in the source image. A few photography choices make difficult boundaries easier to identify.
Motion blur and missed focus blend foreground and background pixels together. If possible, use a photo where hair, shoulders, product edges, and clothing seams remain visibly defined.
A subject wearing dark clothing against an equally dark background is harder to separate than the same subject against a visibly different background. You do not need a studio backdrop, but clear visual separation helps.
Heavily compressed screenshots and repeatedly shared images can contain block artifacts around edges. Starting from the original camera image or a higher-quality copy gives the model more information to work with.
Veils, glass, smoke, reflections, flyaway hair, fur, lace, and motion blur are naturally ambiguous. Zoom in after removal and check those areas before using the result in a final design or document.
The current image path remains browser-side.
Source photos and generated cutouts stay in browser memory while you work. Model/runtime assets can still be downloaded by the browser from their configured provider, but FlytheBG does not intentionally attach your source image to those asset requests.
AI segmentation is an estimate, not a guarantee.
No automatic remover can guarantee a perfect result for every image. Difficult foregrounds can be clipped, background fragments can remain, and fine translucent details can be simplified. Check important results at full size before publishing or printing them.
Need physical photo sizes?
Use the Passport Photo Maker to frame the result, set real-world dimensions, choose a photo background, and build a white print sheet. Document authorities can have additional rules beyond image size.
Open Passport Photo Maker ↗ Remove BG FAQ
Common questions before you process an image.
These answers describe the current FlytheBG workflow rather than promising results the model cannot guarantee.
Can I use portrait, landscape, square, or panoramic images?+
Yes. FlytheBG preserves the source aspect ratio instead of forcing the image into a square. Extremely large files are still limited by browser memory and the configured upload-size guard.
What kind of images work best?+
Clear photos with a visible subject boundary usually produce the most reliable result. Strong lighting, useful resolution, and some contrast between the foreground and background help the model separate fine details.
Why can hair, fur, glass, or motion blur be difficult?+
Those areas contain partially transparent or ambiguous pixels. A segmentation model must estimate which pixels belong to the foreground, so very fine strands, reflections, transparent objects, smoke, and heavy blur can need manual checking.
Does FlytheBG send my photo to an image-processing server?+
The current production remover runs IMG.LY in the browser. The page does not intentionally send source or generated image bytes to a FlytheBG inference server or image database.
Why is the first run sometimes slower?+
The browser may need to download the smaller quantized model and runtime assets before processing the first image. Compatible browsers try WebGPU first, and CPU/WASM is the automatic fallback.
What file do I download?+
The background-removal workflow exports a PNG so the removed background can remain transparent. Applications that do not display transparency may show the transparent area as white, black, or a checkerboard preview.