# AI background remover for clean cutouts

> Remove backgrounds from JPG, PNG, or static WebP with AI. Get a transparent PNG at the original size; up to 10 images, 10 MB each, 24 hours.

**AI Background Remover - Remove Image Backgrounds Online | docsmall**

- Canonical HTML: https://docsmall.com/en/image-remove-bg
- Markdown: https://docsmall.com/en/image-remove-bg.md
- Language: en
- Updated: 2026-11-18
- Limits: 10 images, 10 MB each, JPEG/PNG/static WebP, 5000 px longest side; server-side AI cutout, results kept for 24 hours

Let an AI model remove the background from JPG, PNG, and static WebP images, keep the subject, and return a transparent PNG.

docsmall removes image backgrounds with an AI model: up to 10 images per batch, 10 MB per image, and 5000 px on the longest side. Click “Remove background” after the uploads finish to create the tasks. The output is a transparent PNG with the same width and height as the original, and results are kept for 24 hours.

## Cost · Processing · Input limits · Output

- **Cost:** Daily free credits; 1 credit per image
- **Processing:** Server-side; images are uploaded
- **Input limits:** 10 images, 10 MB each, 5000 px longest side
- **Output:** Transparent PNG at the original size

This is a fully automatic AI cutout tool for product photos, portraits, avatars, design assets, and presentation images: the model decides what is subject and what is background, then returns a transparent PNG you can place on any backdrop. It does not offer brush, pen, or mask refinement, and it cannot process animation, video, or images whose longest side exceeds 5000 px.

- Click “Remove background” after the uploads finish to create the tasks.
- Transparent PNG: the result keeps the original width and height, with a transparent background and soft alpha edges.
- Batch and download: switch thumbnails to compare the original with the cutout, download one file, or pack at least two successful results into a ZIP.

## How to remove an image background

Choose up to 10 images, let the model remove each background automatically, then inspect the edges and download the transparent PNG. Processing happens on the server, so images are uploaded.

1. **Choose images** — Click the upload area, drop files into it, or focus it and press Enter or Space. Add up to 10 JPG, PNG, or static WebP images, 10 MB each, with a longest side of no more than 5000 px. Animated WebP, damaged files, and files whose content does not match the extension are rejected.
2. **Remove background** — Click “Remove background” after the uploads finish to create the tasks. Uploading starts immediately, and each image enters the queue when its upload finishes; at most two images are processed at the same time. Thumbnails show upload progress, processing, or failure. Keep the page open until the status reads “Background removed”.
3. **Compare and download** — Click a finished thumbnail and the result wipes in from left to right over the original. Use “View original” and “View result” to inspect the edges, then download the current image. When at least two images succeed, you can pack them into a ZIP. Each file is named after the original stem with a .png extension; expired results must be processed again.

## Guide to automatic background removal

How AI cutouts work, which images and limits apply, what the transparent PNG contains, where images are processed and how long results last, and which subjects are a poor fit for automatic removal.

### What does this tool return?

It separates the subject from the background, keeps the subject, and makes the background transparent. The output is a PNG file with the same width and height as the original and a transparent background.

Unlike manual cutouts, you do not draw a selection or mask: the model judges which areas belong to the subject and produces transparency at the original size. Edges therefore have soft transitions, and hair, fur, and semi-transparent materials may keep some background color.

The tool handles still images only. It does not create vector paths and does not replace the background, add shadows, or reposition the subject. If you need a white background or a JPG, first create the transparent PNG here, then use [Image to JPG](https://docsmall.com/en/convert-to-jpg) to composite a background.

- Output: transparent PNG with the same width and height as the original.
- Input: JPG, PNG, or static WebP; up to 10 images, 10 MB each, 5000 px longest side.
- No manual refinement, background replacement, or vector export.

### How does the AI decide what is subject and background?

The model looks at a smaller version of the image to judge which areas belong to the subject, then produces transparency at the original size. When scaled back up, edges become soft transitions rather than hard lines.

If the original already has transparent areas, the tool first composites them onto white before recognition, then combines the result with the original transparency. Existing transparent areas are not filled in just because the model guessed subject there.

The process is deterministic: the same image usually produces the same result with the same model version, while different images, compression levels, and subject complexity produce visibly different outcomes. The model does not know that something is “a product” or “a person”; it segments based on visual patterns seen during training.

- Judges the subject area first, then produces transparency at the original size.
- Existing transparent areas are multiplied with the prediction instead of being filled in.
- Results depend on subject-background contrast, edge complexity, and compression.

### Which images are supported, and why these limits?

Up to 10 JPG, PNG, or static WebP images per batch. Each can be up to 10 MB with a longest edge of 5000 pixels. The browser checks format, size, and dimensions first, and the server checks again after receiving the file.

The 5000-pixel limit keeps memory use and processing time per image under control: a 5000 × 5000 image takes about 100 MB once expanded. Images above the limit are not silently shrunk; resize them below 5000 pixels first with [Image Resize](https://docsmall.com/en/image-resize).

GIF, BMP, TIFF, HEIC, AVIF, and SVG are outside the scope of this page, and animated or multi-frame WebP files are rejected because the model processes one still frame at a time. Corrupt, truncated, or mislabelled files are also rejected.

- Up to 10 images per batch; 10 MB each and 5000 px on the longest side.
- Only JPG, PNG, and static WebP; animation and multi-frame files are rejected.
- Both the browser and the server validate format, size, and dimensions.

### What is special about the transparent PNG?

The output is always a PNG with the same width and height as the processed original. It is not cropped or scaled to fit a canvas. If the original has an EXIF orientation, the server corrects it first, so the output dimensions can be swapped relative to the values stored in the file.

PNG uses lossless compression, which suits further editing and layering; a PNG of the same content is usually larger than JPG, and transparent areas add to file size. For a smaller file, use [Image Compress](https://docsmall.com/en/image-compress) or convert to a JPG with a background for its final use.

The output does not keep EXIF, comments, or other metadata from the original. If the source already had transparent areas, they are preserved and combined with the predicted transparency; the background becomes transparent only where the source was fully opaque.

- Always PNG, with the same width and height as the original.
- Lossless, so files are usually larger than JPG.
- EXIF and comments are dropped; existing transparency is preserved.

### How are images uploaded and processed?

Click “Remove background” after the uploads finish to create the tasks. The page shows upload and processing progress, and you can download the result when it is ready. No account is required.

Click “Remove background” after the uploads finish to create the tasks. At most two images are processed at once, and the rest stay in the waiting state. A failed image does not block the others, and failed tasks are not retried automatically; clear them and choose the files again.

Results are kept for 24 hours. Reopening the page does not restore a previous queue, and downloading an expired result prompts you to process the image again.

- Images are uploaded and processed by an AI model on the server.
- Click “Remove background” after the uploads finish to create the tasks.
- Results are kept for 24 hours and failures are not retried automatically.

### Automatic or manual cutout?

Automatic removal suits batches of images with a clear subject and a relatively simple background: product shots on white, ID-photo background swaps, avatars, and illustration assets. It handles up to 10 images at once without tracing each outline by hand.

Manual cutouts suit single images with extreme edge requirements: wedding dresses, glassware, complex hair, translucent objects, or scenes where an exact shadow must be preserved. This page has no brush, lasso, or mask, so you cannot refine an edge after processing.

In practice, you can run a batch here for a first pass and refine only the few images that need it in an image editor. You can also use [Image Crop](https://docsmall.com/en/image-crop) to tighten the frame around the subject before removing the background, which reduces distracting background detail.

- Clear subject, simple background, many images: prefer automatic removal.
- Extremely complex edges, few images: use a manual tool with masks.
- Crop before removing the background to reduce irrelevant detail.

### Where does automatic removal work best?

E-commerce and catalogues: separate products from the shooting backdrop and place them on a consistent colour, scene, or store template. Results are more consistent when a batch comes from the same shoot with similar lighting.

Portraits and avatars: remove the background from ID photos, social avatars, or team pages before compositing a solid or brand background. Hair edges deserve a close look, especially dark hair against a light background.

Design and content: lift illustrations, icons, or parts of a product out of a screenshot for posters, slides, video covers, or web banners. A transparent PNG can sit on any background without further masking.

- Product photos, catalogue images, and advertising assets.
- ID photos, avatars, and team pages.
- Illustrations, icons, slides, and web imagery.

### Are images uploaded to a server?

Yes. Background removal runs on an AI model on the server, images are uploaded to docsmall storage, and results are stored there too. Your browser does not remove the background locally.

No account is needed, and results are deleted after 24 hours. Do not upload ID cards, bank cards, medical records, unpublished contracts, or other sensitive images unless you accept that they will be uploaded and processed on a server.

Uploads and downloads use HTTPS. The page does not write image content to local storage, and closing the page does not delete an uploaded task before its expiry time.

- Images are uploaded; the cutout runs on the server.
- Results are kept for 24 hours and no account is required.
- Do not upload ID cards, contracts, or other sensitive images.

### Limitations, compatibility, and caveats

Automatic removal is not guaranteed to succeed on the first try. When subject and background have similar colours, when the background is busy, or when the photo is backlit, motion-blurred, partially occluded, or contains overlapping subjects, edges can retain background pixels or lose detail.

Hair, fur, feathers, tulle, glass, smoke, flames, and clear liquids are the hardest materials. The model may keep too much background or cut away a translucent area entirely. Product shadows and reflections are usually treated as part of the background and removed.

There is no manual edge repair, undo, or local re-run; if you dislike the result, the only option is to try a clearer image. Images beyond 10 files, 10 MB, or 5000 px are rejected, and HEIC files must be converted to JPG or PNG before upload.

- Complex edges, transparent materials, shadows, and reflections may fail.
- No manual edge repair or local re-run.
- Images beyond the count, size, or dimension limits are rejected.

## JPG, PNG, and WebP cutout differences

All three inputs produce a transparent PNG, but the source transparency, compression artefacts, and file size affect the result and what you can do next.

| Input | Typical traits | After removal | Advice |
| --- | --- | --- | --- |
| JPG | Lossy compression, usually no transparency, common for photos and product shots. | Transparent PNG output; compression noise can leave slight colour fringing on edges. | Best for photographic subjects. Start from a high-quality original and avoid repeated re-saving. |
| PNG | Lossless compression, sometimes transparent, common for screenshots, icons, and design files. | Transparent PNG output; existing transparency is preserved and multiplied with the model result. | Good for icons, illustrations, and assets whose original transparency must survive. |
| Static WebP | Small files, lossy or lossless, common on the web. | Transparent PNG output; recompression can cost a little edge detail. | Single frames only. Export a still frame from an animated WebP first. |

## Before and after background removal

Three examples show transparent results for a portrait, a product, and a design asset. Drag the slider to inspect the edges.

- Portrait: [Original](https://docsmall.com/images/image-remove-bg/people-before.webp) · [Transparent PNG](https://docsmall.com/images/image-remove-bg/people-after.webp)
- Product: [Original](https://docsmall.com/images/image-remove-bg/product-before.webp) · [Transparent PNG](https://docsmall.com/images/image-remove-bg/product-after.webp)
- Asset: [Original](https://docsmall.com/images/image-remove-bg/material-before.webp) · [Transparent PNG](https://docsmall.com/images/image-remove-bg/material-after.webp)

Examples show typical cutout results. Actual output depends on the source image, subject edges, lighting, and compression.

## 6 ways to get a more usable cutout

These habits reduce edge residue and help you judge whether a result holds up at its final size.

### Start with a sharp, evenly lit original

The clearer the contrast between subject and background, the easier the separation. Avoid backlighting, heavy noise, strong colour casts, and motion blur; use even lighting when shooting with a phone.

### Use the highest resolution under 5000 px

More resolution means more edge detail. Do not shrink a large photo below 5000 px just to fit the limit and then upscale it again; use a sharp original that already fits the limit.

### Crop before removing the background

Use [Image Crop](https://docsmall.com/en/image-crop) to remove irrelevant background and empty margins. A larger subject gives the model more context at the edges and reduces the upload size.

### Inspect at 100% on different backdrops

Place the transparent PNG on dark, light, and brand-coloured backgrounds. Hair, white clothing, and pale products are the first places where residual background shows up.

### Keep the original file

An automatic result is not guaranteed to satisfy you on the first attempt. Keep the source, retry with a sharper version if needed, or finish the edges in an image editor.

### Compress the transparent PNG before publishing

Transparent PNGs are usually larger than the source. For web or store backends, use [Image Compress](https://docsmall.com/en/image-compress) to reduce the size, then [resize](https://docsmall.com/en/image-resize) to the platform’s required dimensions so it is not recompressed after upload.

## Background removal FAQ

### How is this tool priced?

A free allowance is granted every day, and each image costs 1 credit. You can buy more credits when they run out; paid credits last 24 hours. Credits are charged when a task is created, but a failed task is refunded automatically, so you are not charged for it. See [Free credits and paid top-ups](https://docsmall.com/en/credits).

### Are my images uploaded, and how long are they kept?

Yes. Background removal happens on the server: images are uploaded to docsmall storage, and results are deleted after 24 hours. Do not upload ID cards, contracts, or other sensitive images.

### Which formats and sizes are supported?

JPG, PNG, and static WebP; up to 10 images per batch, 10 MB each, with a longest side of no more than 5000 px. GIF, BMP, TIFF, HEIC, AVIF, SVG, and animated WebP are not supported.

### Do output dimensions change?

No. The PNG keeps the original width and height. If the source has an EXIF orientation, the server applies it first, so the output width and height can be swapped relative to the values stored in the file metadata.

### Why is the output PNG instead of JPG?

Transparency requires PNG; JPG does not support it. If you need a JPG, use this page to get a transparent PNG first, then use [Image to JPG](https://docsmall.com/en/convert-to-jpg) to place it on white or another background.

### What is the checkerboard in the preview?

The checkerboard only indicates transparency on screen and is not written into the output. When you open the downloaded PNG in software that supports transparency, the background is still transparent.

### Can it cut out hair, fur, and tulle cleanly?

The model tries to keep semi-transparent edges, but strands of hair, fur, feathers, and tulle can still keep background colour or lose detail. The problem is most visible with dark hair on a light background or in backlit photos; inspect at 100%.

### How many images can I process, and how long does it take?

Up to 10 images per batch, with at most two processed at once and the rest queued. Time per image depends on its dimensions and server load. The page updates status automatically, so you do not need to refresh.

### Why did it say no subject was found?

The image may be almost a single colour, the subject may have too little contrast with the background, or there may be no clear foreground. Try an image with a larger subject and clearer edges.

### What should I do if removal fails or times out?

Failed tasks are not retried automatically. Clear the failed item, check that the file is not damaged and the format is supported, then choose it again. If it still fails, try a smaller or simpler image.

### Can product shadows or reflections be kept?

Usually not. The model treats shadows and reflections as part of the background and removes them. If they matter, composite them separately in an image editor or use a tool with manual masking.

### What happens with glass, smoke, or flames?

These materials have no clear solid edge. The model may keep too much background or remove a translucent area entirely. Automatic removal is not suitable for precise separation of such subjects.

### Can I download an expired result?

No. Results are kept for 24 hours, and an expired download asks you to process the image again. Download files you need in time, or keep the original so you can upload it again.

### How do I handle HEIC photos from a phone?

This page does not accept HEIC. Export the photo as JPG or PNG on the phone or computer first, keep the original resolution, and avoid excessive compression before uploading.

## Related image tools

After the cutout, use these tools to finish the crop, dimensions, and file size.

- [Image Compress](https://docsmall.com/en/image-compress) — A transparent PNG is usually larger than the source; compress it before publishing to the web or a store backend.
- [Image Crop](https://docsmall.com/en/image-crop) — Crop around the subject after removing the background to get a tighter product image or avatar.
- [Image Resize](https://docsmall.com/en/image-resize) — Resize the transparent PNG to the exact dimensions a platform requires so it is not reprocessed after upload.

## Choose or drop images to remove the background

- Up to 10 JPG, PNG, or static WebP images, 10 MB each, longest side no more than 5000 px.
- Images are uploaded and processed by an AI model on the server. Each image is processed separately.
- Output is a PNG with a transparent background; pixel dimensions match the original.
- Image queue

## Clear files / Download

- **Clear files** / **Download** / **Download all**
- View original / View result

- 2 images selected
- Finished 1 of 2
- All 2 images finished
- Uploading 42% · Ready to remove background · Removing background… · Background removed

## English

[中文](https://docsmall.com/image-remove-bg) · [**English**](https://docsmall.com/en/image-remove-bg) · [Français](https://docsmall.com/fr/image-remove-bg) · [Español](https://docsmall.com/es/image-remove-bg) · [Italiano](https://docsmall.com/it/image-remove-bg) · [Deutsch](https://docsmall.com/de/image-remove-bg) · [Bahasa Indonesia](https://docsmall.com/id/image-remove-bg) · [Português](https://docsmall.com/pt/image-remove-bg) · [日本語](https://docsmall.com/ja/image-remove-bg) · [한국어](https://docsmall.com/ko/image-remove-bg) · [ไทย](https://docsmall.com/th/image-remove-bg) · [Tiếng Việt](https://docsmall.com/vi/image-remove-bg) · [العربية](https://docsmall.com/ar/image-remove-bg) · [Русский](https://docsmall.com/ru/image-remove-bg)

---

[docsmall](https://docsmall.com/en) · [HTML](https://docsmall.com/en/image-remove-bg) · [llms.txt](https://docsmall.com/llms.txt) · [sitemap.xml](https://docsmall.com/sitemap.xml)
