---
title: "How to remove a photo background for free · Cutshed"
description: "Remove a photo background in the browser with U²-Netp and download a transparent PNG. The photo is not uploaded. The long edge stays at most 1600 pixels."
canonical: "https://cutshed.com/blog/how-to-remove-the-background-from-a-photo/"
---

# How to remove a photo background for free · Cutshed

Remove a photo background in the browser with U²-Netp and download a transparent PNG. The photo is not uploaded. The long edge stays at most 1600 pixels.

- Products
- Studio
- Video Game Clipper
- Stream Clipper
- Week of Shorts
- Editor
- Free tools
Audio
- Vocal remover
- Audio trimmer
- Ringtone maker
- Audio converter
- Volume booster
- Noise remover
- BPM and key
- Pitch and speed
- Clip sound effects
- AI audio maker
Video
- Video editor
- Video compressor
- Video trimmer
- Crop and resize
- Frames and thumbnails
- Mute video
- Video speed
- Merge clips
- Video to MP3
- Video to GIF
- Video splitter
- Week of Shorts
Image
- Background remover
- Image upscaler
- Image converter
- Thumbnail maker
- AI emoji maker
Subtitles & text
- Subtitle generator
- Subtitle translator
- Timing shifter
- Transcript to post
- Resources
- Menu
Resources
Audio
- Vocal remover
- Audio trimmer
- Ringtone maker
- Audio converter
- Volume booster
- Noise remover
- BPM and key
- Pitch and speed
- Clip sound effects
- AI audio maker
Video
- Video editor
- Video compressor
- Video trimmer
- Crop and resize
- Frames and thumbnails
- Mute video
- Video speed
- Merge clips
- Video to MP3
- Video to GIF
- Video splitter
- Week of Shorts
Image
- Background remover
- Image upscaler
- Image converter
- Thumbnail maker
- AI emoji maker
Subtitles & text
- Subtitle generator
- Subtitle translator
- Timing shifter
- Transcript to post
- Home
- / Blog
- / How to remove the background from a photo for free
How to
# How to remove the background from a photo for free
Published November 20, 2026 .
Open Background remover, drop a still, and click Remove background. U²-Netp runs in the browser on a 320 by 320 input and writes one transparent PNG. The long edge of that PNG is at most 1600 pixels. Files larger than 25 times 1024 times 1024 bytes are refused. The photo is not uploaded.
The tool for this is Background remover . It takes a still you already have and writes a PNG whose alpha channel is a subject mask. The pixels stay in the tab. Use a photo you are allowed to post.
This page cuts one subject out of one picture. It does not paint a new color behind that subject, and it does not make a video matte. A hosted cutout that you upload is a different product. The on-device comparison is a free Remove.bg alternative .
## Steps
- Drop a still Open Background remover and drop the file. The picker accept string is image/*,.png,.jpg,.jpeg,.webp,.gif,.avif. The hint under the picker says a photo with a clear subject, and it names PNG, JPG, or WebP. The page decodes with createImageBitmap. It does not contain the HEIC refusal used by Image converter. A file this browser cannot decode fails at that bitmap step. A file larger than 25 times 1024 times 1024 bytes is refused when you click Remove background, and the message says to keep it under 25 MB. A file of exactly that many bytes is not over the limit. There is no duration check, because this is a still.
- Read the note before you cut The note says the cutout uses U²-Netp, a 4.4 MB salient-object model under Apache-2.0. It says the mask is a general subject mask, not a studio key, so hair and glass can be rough. It says RMBG weights are not used because their license is non-commercial. It also says the first run downloads the model, at the byte size in the vendor manifest, plus the ONNX runtime if that runtime is not cached. The tool page names the weights as the small U²-Net variant from the xuebinqin/U-2-Net project, with ONNX weights matching the rembg u2netp asset. RMBG-1.4 is not in this page.
- Click Remove background The button reads Remove background, then Cutting out…, and it is disabled while the job runs. The status line is the WASM backend. On the first run the progress text is Downloading the cutout model, with the loaded and total sizes. That download is the weight file at /models/u2netp.onnx. The session is created with executionProviders set to wasm only. The runtime file is ort-wasm-simd-threaded, and the thread count is the smaller of 4 and the machine's hardwareConcurrency (or 1 if that count is missing). A later photo in the same tab reuses the session, so it does not start that weight download again. If the page is not cross-origin isolated, the session throws and the message names Cross-Origin-Opener-Policy same-origin and Cross-Origin-Embedder-Policy require-corp. The photo itself is not uploaded.
- Check the checkerboard, then download the PNG When the job finishes, the status says PNG ready and the size of the new file. The cutout image appears under the button, on a checkerboard drawn by the page style. Its alt text starts with the word Cutout. The checkerboard is not painted into the pixels. Click Download PNG. The file name is the stem of the name you dropped, plus .png. The stem drops the extension, replaces characters outside letters, digits, underscore, dot, and hyphen, collapses repeated hyphens, and keeps at most 80 characters. If the file has no name, the stem is cutout. The download is a separate click. Remove background does not save the file by itself. Closing the tab drops the cutout image. The file you dropped is unchanged.
## What Background remover will and will not do
## What the mask changes, and what it leaves alone
The page draws the photo into a working canvas whose long edge is at most 1600 pixels. The scale is the smaller of 1 and 1600 divided by the longer side, so a smaller photo is not enlarged. Width and height are rounded, and each is at least 1 pixel. Those working pixels are the color in the PNG.
The same bitmap is also stretched into a 320 by 320 canvas. That stretch is not letterboxed. Each pixel is normalized with ImageNet mean 0.485, 0.456, 0.406 and standard deviation 0.229, 0.224, 0.225, then sent as one float32 tensor of shape 1, 3, 320, 320. U²-Netp returns a 320 by 320 mask. The page samples that mask with nearest pixels back onto the working image: for each output pixel it takes floor of the 320-grid position, clamps the mask value to the range 0 through 1, and writes round of that value times 255 into the alpha channel. Red, green, and blue stay the working-image colors. There is no blur pass after the sample.
The file is one PNG from canvas.toBlob with type image/png. There is no quality slider and no format menu. A GIF or any other multi-frame file becomes that one bitmap. The download is not an animation.
## What this page will not fix
There is no brush, no eraser, and no background-color control. The tool copy says a halo or a hole is not fixed by running the cut a second time, and that those pixels are a job for an editor or a cleaner backdrop. This page does not include that paint tool. Hair and glass are the cases the note calls rough. A dark subject on a dark couch is the case the tips say the mask gives up on. Cropping in close matters because the model only sees 320 pixels on a side.
The page does not replace the background with a color, a gradient, or a generated scene. Transparent pixels stay transparent. Image converter can flatten that PNG to JPG, and JPG paints transparency onto white. Quality there opens at 0.8 and runs from 0.4 to 1. The longest edge opens at 1920 and can be 640, 1280, 1920, or 2560. Smaller sources are not enlarged. HEIC is refused on the converter. The same 25 MB file cap applies there. Use the converter when a form wants a flat JPG. Use the PNG when the next step needs the hole.
It is not a video tool. The editor does not remove a video background. A generated picture behind a subject is a Studio job: it spends credits and the prompt leaves the machine. Background remover does not. One photo at a time is the whole batch. There is no API on this page.
## Use the PNG on a thumbnail
Thumbnail maker is a 1280 by 720 canvas. An optional photo covers the frame and then a dark scrim is painted over it. That scrim is a flat color, so a transparent PNG dropped there does not keep a see-through hole on the thumbnail. The export is a flat PNG named thumbnail.png. The walkthrough is how to make a YouTube thumbnail for free .
Image upscaler multiplies each side by four after capping the long edge at 512 pixels, and the output alpha is set to 255, so it is not a way to enlarge this cutout and keep the hole. Run the cutout at the size you need. The background page already capped the long edge at 1600 before the mask was applied.
## Do this part in Cutshed
- Background remover U²-Netp in the browser. One transparent PNG. Long edge at most 1600 pixels. Files over 25 MB are refused. Nothing is uploaded.
- Image converter JPG, PNG, WebP, or AVIF. JPG flattens transparency onto white. HEIC is refused. The same 25 MB cap.
- Thumbnail maker A 1280 by 720 canvas and an optional photo under a dark scrim. Export is a flat PNG named thumbnail.png.
- Image upscaler Four times each side after a 512-pixel long-edge cap. The output alpha is opaque.
## FAQ
- How do I remove a photo background for free without uploading it?
Open Background remover, drop the still, and click Remove background. There is no account. U²-Netp runs in the tab. The weights download once. The photo is not sent anywhere.
- Which files are accepted, and what is the size limit?
The picker accepts image files plus png, jpg, jpeg, webp, gif, and avif. The hint names PNG, JPG, or WebP. Decoding uses createImageBitmap, so a format this browser cannot decode fails there. This page does not add the converter's HEIC refusal. A file larger than 25 times 1024 times 1024 bytes is refused, and the message says to keep it under 25 MB.
- What do I download?
One PNG. Alpha is the U²-Netp mask. Red, green, and blue are the working image, with the long edge at most 1600 pixels. Download PNG uses the original file stem plus .png. The checkerboard behind the cutout image is not in the file.
- Why does a large photo come back smaller?
Before the mask, the long edge is scaled to at most 1600 pixels. Smaller photos are not enlarged. The model itself only sees a 320 by 320 stretch of that bitmap, and the page samples that mask back with nearest pixels.
- Will it keep hair, and can I paint the edge?
The page says sometimes, as a soft edge, and that fine hair on a busy background is what this small model misses. There is no brush on this page. Running the button again does not repair a halo or a hole.
- Can I put a new color or a new scene behind the subject?
Not here. The PNG is transparent where the mask is low. Image converter can flatten it to JPG, which fills those pixels with white. A generated scene is a Studio job, which spends credits and uploads the prompt. This remover does neither.
Files stay on this device.
Free tools never upload the file you open. Studio is the only paid part of the site, and the only place a prompt is sent to a video model.

## Links

- [C Cutshed](/)
- [Studio](/studio/)
- [Video Game Clipper](/video-game-clipper/)
- [Stream Clipper](/stream-clipper/)
- [Week of Shorts](/stream-to-shorts/)
- [Editor](/editor/)
- [Free tools](/#tool-grid)
- [Vocal remover](/vocal-remover/)
- [Audio trimmer](/audio-trimmer/)
- [Ringtone maker](/ringtone-maker/)
- [Audio converter](/audio-converter/)
- [Volume booster](/volume/)
- [Noise remover](/noise-remover/)
- [BPM and key](/bpm-key/)
- [Pitch and speed](/pitch-speed/)
- [Clip sound effects](/clip-sound-effects/)
- [AI audio maker](/ai-audio/)
- [Video compressor](/video-compressor/)
- [Video trimmer](/video-trimmer/)
- [Crop and resize](/video-crop/)
- [Frames and thumbnails](/video-frames/)
- [Mute video](/mute-video/)
- [Video speed](/video-speed/)
- [Merge clips](/merge-clips/)
- [Video to MP3](/video-to-mp3/)
- [Video to GIF](/video-to-gif/)
- [Video splitter](/video-splitter/)
- [Background remover](/background-remover/)
- [Image upscaler](/image-upscaler/)
- [Image converter](/image-converter/)
- [Thumbnail maker](/thumbnail-maker/)
- [AI emoji maker](/ai-emoji-maker/)
- [Subtitle generator](/subtitles/)
- [Subtitle translator](/subtitle-translator/)
- [Timing shifter](/subtitle-shifter/)
- [Transcript to post](/transcript/)
- [Blog](/blog/)
- [Compare](/compare/)
- [Developers](/developers/)
- [Seedance 2.5](/seedance-2-5/)
- [Privacy](/privacy/)

