Background Remover
Remove image backgrounds with an in-browser AI model download. Supports PNG, JPG, and WebP inputs up to 10 MB, then exports a transparent PNG.
Upload a photo and the tool automatically removes the background using in-browser AI. Download the result as a transparent PNG, ideal for product shots and headshots.
One-time setup is waiting
The background removal model download is about 320 MB. After it loads, the selected image is processed by the page in your browser.
Model Download
About 320 MB on first use
Browser Processing
Runs after model files load
PNG Output
Transparent result when processing succeeds
Current Processing Notes
- Input: PNG, JPG, JPEG, or WebP up to 10 MB.
- Model files: About 320 MB on first use.
- Processing: One image at a time in a compatible browser.
- Output: Transparent PNG when processing succeeds.
Background removal powered by withoutBG Focus, an open-source matting pipeline. Visit withoutbg.com.
How AI Background Removal Works
This tool runs the withoutBG Focus pipeline, a four-stage matting system. It is not a single model but a sequence of four specialised neural networks, each one handing its output to the next. The four stages help separate subjects from backgrounds, then refine soft edges such as hair, fur, and fine fabric.
Stage 1, depth estimation. Depth Anything V2 (ViT-Small) scans the image at 518-pixel resolution and produces a depth map, one number per pixel telling the next stage how far each part of the scene is from the camera. Depth is a strong prior for separating a subject from its background because subjects sit at a different distance from the camera than what is behind them.
Stage 2, saliency segmentation. ISNet looks at the image at 1024-pixel resolution and generates a coarse saliency mask, a first pass at "what is the subject." This is the broad-strokes cutout before any edge refinement.
Stage 3, alpha matting. A purpose-trained matting network takes the RGB image, the depth map, and the ISNet saliency mask together as a 5-channel input and produces a proper alpha matte with soft edges. This is where hair strands and fur tips get their transparency values, values between 0 (fully background) and 1 (fully subject), instead of a hard binary cut.
Stage 4, refinement. A refiner network takes the RGB image plus the coarse alpha matte from stage 3 and cleans up the edges at higher resolution, fixing jagged boundaries and preserving fine detail that the lower-resolution matting stage could not see.
All four models run in your browser using WebAssembly with ONNX Runtime Web. On first visit the four ONNX files, roughly 320 MB in total, download from the iForge Apps CDN. Later visits may load cached model files faster, depending on browser storage, cache settings, and whether the cached files are still available.
Model notes. withoutBG publishes model and dataset notes at withoutbg.com. Review the upstream project details if licensing, training data, or commercial-use permissions matter for your workflow.
After model files load, the selected image is read by browser APIs and processed in the page. The reviewed implementation does not call an iForge Apps upload endpoint for the selected image. Test the flow with a non-sensitive image first if your file is confidential.
Current Background Remover Notes
These notes describe the current Forge BG page. Check your output before using it for marketplaces, identity documents, client work, or confidential images.
| Feature | iForge Apps | Notes |
|---|---|---|
| Input limit | 10 MB | PNG, JPG, JPEG, or WebP image files |
| Model download | About 320 MB | Required before first processing run |
| Processing | Browser page | Runs with ONNX Runtime Web after model files load |
| Output | PNG | Transparent PNG when processing succeeds |
| Batch processing | One image | Process one file at a time |
| Caching | Browser-dependent | Later visits may be faster if model files remain cached |
What this means for you: Use this page for single-image background removal where a transparent PNG is useful. For sensitive or high-stakes files, test with a non-sensitive image and inspect the result carefully.
What Works Well (And What Doesn't)
| Subject Type | Result Pattern | Notes |
|---|---|---|
| Person (clear background) | Usually strong | Clean edges, good hair detail, the model's strongest use case |
| Product on white/plain | Usually strong | Often clean for e-commerce when product edges are distinct |
| Pet/animal | Very good | Fur edges may need slight touch-up on fluffy breeds |
| Food photography | Very good | Works well on plates/bowls; sauce drips may be clipped |
| Person (busy background) | Good | Main body clean; complex hair-on-foliage is harder |
| Multiple subjects | Good | Preserves all foreground elements, crop first if you only want one |
| Vehicle/car | Good | Body is clean; reflections on glass/chrome can be tricky |
| Text/logo on background | Variable | Works if text is the clear foreground; thin strokes may clip |
| Semi-transparent objects | Challenging | Glass, smoke, veils, and water splashes lose transparency |
| Subject matches background | Challenging | White product on white table, black cat on dark sofa, not enough contrast |
What this means for you: The golden rule is contrast. The more your subject stands out from its background in colour, brightness, or texture, the cleaner the result. If you're photographing products specifically for background removal, use a background colour that contrasts with the product, a blue cloth behind a white mug, for example.
Platform Image Requirements
After removing the background, you'll need to export at the right size and format for your destination. This table covers every major e-commerce platform, social media channel, and design tool, search for yours.
| Platform | Recommended | Background |
|---|---|---|
| Amazon | 2000 × 2000 | Pure white (RGB 255,255,255) |
| eBay | 1600 × 1600 | White or light grey |
| Etsy | 2700 × 2025 | Flexible, styled shots allowed |
| Shopify | 2048 × 2048 | Consistent across all products |
| Depop | 1280 × 1280 | Clean, simple (white preferred) |
| Vinted | 1080 × 1080 | Clean background, no clutter |
| Instagram Post | 1080 × 1350 | Any, transparent for stories |
| Instagram Story | 1080 × 1920 | Transparent PNG for layering |
| Facebook Post | 1200 × 630 | Any |
| LinkedIn Post | 1200 × 627 | Professional, clean |
| Twitter/X Post | 1600 × 900 | Any |
| TikTok | 1080 × 1920 | Any, green screen effect built-in |
| YouTube Thumbnail | 1280 × 720 | High contrast recommended |
| Pinterest Pin | 1000 × 1500 | Clean, bright backgrounds often perform well |
| Canva Template | 3000 × 3000+ | Transparent PNG for layering |
| Figma/Sketch | 2x or 3x display size | Transparent PNG |
| PowerPoint/Keynote | 1920 × 1080 | Transparent PNG for slide layering |
| Google Merchant | 800 × 800+ | White or transparent |
| Passport Photo (UK) | 600 × 750 | Plain light grey or cream |
| ID Photo (US) | 600 × 600 | Plain white |
What this means for you: Amazon is the strictest, main images MUST have pure white backgrounds (RGB 255,255,255). For social media, transparent PNGs give you the most flexibility for layering in design tools. UK passport photos need light grey, NOT white, the passport office rejects pure white backgrounds. Use Image Resizer to hit exact pixel dimensions after removal.
Output Format Guide
After removing the background, you need to save in the right format. The wrong choice either loses transparency or creates unnecessarily large files.
| Format | Typical File Size | Transparency | Good For | Quality |
|---|---|---|---|---|
| PNG (transparent) | Large (2-10 MB) | Yes | Layering in designs, e-commerce white backgrounds, social media stories | Lossless format |
| PNG (white background) | Large (2-10 MB) | No | Amazon product photos, passport photos, printing | Lossless format |
| JPEG | Small (100 KB-2 MB) | No | Email, web pages, social media posts, general sharing | Lossy, slight quality loss at high compression |
| WebP | Smallest (50 KB-1 MB) | Yes | Websites, blogs, anywhere file size matters most | Lossy or lossless, good size-to-quality balance |
What this means for you: Save as PNG if you need the transparent background (for layering in designs, Canva, PowerPoint). Save as JPEG if you're going to put a solid colour behind it anyway and want a smaller file. WebP can balance transparency and smaller files, but not every platform accepts it yet. Convert between formats with Image to PNG or Image to WebP.
Worked Example: E-Commerce Product Photos
James sells handmade candles on Etsy and Amazon. He photographs products on his kitchen table with a smartphone. Here's his workflow using free tools:
- 1.Photograph with contrast. James places candles on a blue tea towel. The colour difference between the cream candle and blue cloth makes AI removal much cleaner than shooting on a white surface.
- 2.Remove background. Uploads each photo here. Processing takes 3 seconds per image. Downloads as transparent PNG.
- 3.Amazon version. Uses Image Resizer to create a 2000×2000 square, places the candle on pure white (RGB 255,255,255). Saves as JPEG, Amazon requires white backgrounds.
- 4.Etsy version. Keeps the transparent PNG at 2700×2025 (5:4 ratio). Adds a warm gradient background in Canva for a lifestyle feel. Etsy allows styled images.
- 5.Instagram version. Creates a 1080×1350 (4:5 portrait) with the transparent cutout layered over a branded background. Uses Image Compressor to keep file size under 5 MB for fast upload.
Total workflow: One image at a time, with manual checking before publishing. Professional product photography may still be needed for high-value catalogues.
Tips for Better Results
Maximise contrast
The AI works by detecting edges between subject and background. High contrast usually means cleaner edges. If you're photographing specifically for background removal, choose a background colour that's opposite to your subject. Blue cloth for warm-toned products, dark background for light items.
Use high-resolution source images
Higher resolution gives the AI more pixels to work with at edges, producing smoother cutouts. A 4000px source image gives noticeably better edge quality than a 500px one. Don't upscale before removal, use the original camera resolution.
Light the subject evenly
Harsh shadows on the background can confuse the AI, it might keep the shadow as part of the subject. Natural, diffused light (near a window, or overcast day outdoors) gives the cleanest results. Avoid direct flash.
Crop before processing
If you only want one person from a group photo, crop to just that person first using Image Cropper. The AI keeps all foreground elements, cropping gives you control over which ones.
Handle hair carefully
Wispy hair, curly flyaways, and fine strands are the hardest edges for any AI. Results are usually cleaner when hair is against a solid, contrasting background.
Check edges at 100% zoom
The result may look perfect at thumbnail size but have rough edges when zoomed in. Always check the cutout at full size before using it. Pay particular attention to fingers, hair tips, and any thin protruding elements.
Post-Processing Workflow
Background removal is step one. Here's what typically comes next, depending on your use case:
Resize for destination
Every platform has specific dimensions. Use Image Resizer to match (e.g., 2000×2000 for Amazon, 1080×1350 for Instagram).
Add new background (optional)
For Amazon: add pure white. For social media: use your brand colours. For presentations: keep transparent and layer in PowerPoint/Canva.
Compress for web/email
Transparent PNGs are large. If uploading to social media or email, compress with Image Compressor or convert to WebP with Image to WebP.
Strip metadata (privacy)
Phone photos contain GPS coordinates, camera model, and timestamps. Use Metadata Remover before publishing product images or selfies online.
Common Mistakes
Saving as JPEG after removal
JPEG doesn't support transparency. If you save your transparent cutout as JPEG, the transparent areas become white. Fine for Amazon (they want white), but bad for layering in designs. Use PNG or WebP to preserve transparency.
Using a low-res source image
A 300×200 source image will produce a blurry cutout with blocky edges. The AI needs pixels to work with. Start with the highest resolution available, you can always downscale later, but you can't add detail that isn't there.
Not checking the edges
At thumbnail size, a cutout can look cleaner than it is. Zoom to 100% and check around hair, fingers, and thin elements. A few stray background pixels or clipped subject pixels are obvious when printed or viewed on large screens.
Forgetting about shadows
AI removes the background, including any shadow your subject casts. A product floating with no shadow looks unnatural. For e-commerce, add a subtle drop shadow in your design tool to ground the product.
White product on white background
The AI can't remove what it can't distinguish. Photograph white products against a contrasting colour (blue, grey, even black), remove that background, then place on white. Much cleaner than white-on-white.
Publishing with EXIF data
Phone photos embed GPS coordinates, your name, and device info. If you're posting product photos to marketplaces, strip metadata first with Metadata Remover, especially if you work from home.
Use Cases by Industry
E-commerce & Retail
Product photos on white backgrounds are common marketplace requirements. Consistent backgrounds across your store can help listings look more organised, especially when you inspect and tidy cutout edges before publishing.
Social Media Marketing
Cut out products or people, layer them on branded backgrounds, create before/after comparisons, or make sticker-style graphics. Transparent PNGs work directly in Canva, Instagram story editors, and most design tools.
Real Estate & Property
Replace overcast skies with blue ones, remove distracting elements from property photos, or isolate furniture for virtual staging. Estate agents use background removal more than any industry outside e-commerce.
HR & Corporate
Standardise employee headshots, remove messy office backgrounds and replace with your company's brand colour. One consistent background across all team photos looks polished on the website and LinkedIn.
Education & Presentations
Cut out diagrams, product images, or people for PowerPoint/Keynote slides. Transparent PNGs layer cleanly over any slide background without the ugly white box around the image.
ID & Passport Photos
Remove your background and replace with the required colour (light grey for UK passports, white for US). Check the platform requirements table above, the wrong shade will get rejected.
Related Tools
Image Resizer
Resize cutouts to exact platform dimensions
Image Compressor
Reduce file size for web and email
Image to PNG
Convert to PNG to preserve transparency
Image to WebP
Smallest format with transparency support
Metadata Remover
Strip GPS and personal data before publishing
Image Cropper
Crop to isolate one subject before removal
How to use this tool
Download about 320 MB of model files on first use.
Select a PNG, JPG, or WebP image up to 10 MB.
Wait while the model processes the image, then download the transparent PNG.
Common uses
- Creating transparent product photos for e-commerce listings
- Making profile pictures with clean backgrounds for LinkedIn or CVs
- Preparing images for graphic design, flyers, or presentations
- Removing distracting backgrounds from real estate or food photography
- Isolating subjects for collage, sticker, or social media content creation
Share this tool
Frequently Asked Questions
What does Forge BG need before it can process an image?
How is my selected image handled?
What image formats are supported?
Does the tool have usage limits?
Why does the first visit need to download 320 MB?
Can I remove backgrounds from product photos?
Does it handle hair and fur?
What output format do I get?
Can it handle multiple subjects in one image?
Does it work on mobile?
Why does the background removal leave a halo?
Can I add a new background after removing the old one?
Results are for general informational purposes only and should be checked before use. They are not professional advice. See our Disclaimer and Terms of Service.