Local Background Removal Tool · Runs in Browser
Local Background Removal Tool v2.0.0
100% Local in Browser · Images Never Leave Your DeviceReady
Mode Selection
Auto Cutout
Click Select / Refine
Box Select
Keep
Exclude
Right: remove semi-transparent residue (sofa/table ghost shadows); Left: fill semi-transparent subject to solid (recover incompletely cut pants/skirts). Takes effect instantly, no re-cutout needed
Smooth cutout edges for a natural transition, reducing jagged edges
Image Queue
Marks 0Size --Time --
Ready. Add images to start cutout.
Work Area
Auto mode: add images then click "Cutout Current" for one-click completion
Drag images here, or click to select Also supports Ctrl+V paste · Multiple images at once
Processing…
Processing…
Cutout Result
Cutout results will appear here Supports PNG / WebP / JPEG export
AI ID Photo
make ID photo easier
User Guide
Quick Start (Recommended Workflow)
① Drag in an image (auto cutout is on by default, no button press needed)
② Check the result in the right panel — if satisfied, click "Download Current"
③ Not perfect? Try the "Residue Cleanup / Subject Completion" slider for a quick fix (takes 10 seconds)
④ Still not right? Switch to "Click Select / Refine" mode: click "Exclude" to remove residue, "Keep" to restore missing parts
⑤ Complex images (seated poses, cluttered backgrounds, multiple subjects): select "HD RMBG" model in the left panel for best subject completeness
Two Cutout Modes
Auto Cutout: One-click completion. Three model tiers: Fast (40MB, good for everyday use), Fine (80MB, hair detail), HD RMBG (44MB, more accurate for complex scenes/seated portraits, best subject completeness, recommended).
Click Select: Manually specify the subject — ideal for complex backgrounds and multi-object scenes.
Click Select Operations
Box Select (recommended): Default mode — hold left mouse button and drag a box around the subject. A purple rectangle marker remains after selection (click to delete), visible alongside keep/exclude points
Add Keep Point: Click the "Keep" button to switch mode, then left-click on areas to keep
Add Exclude Point: Click the "Exclude" button to switch mode, then left-click on incorrectly selected areas (e.g., background between hair strands)
Delete Point: Click an existing point (green/red dot) to delete it
Toggle Point Type: Double-click an existing point to toggle between Keep ↔ Exclude
Undo/Clear: Bottom buttons to undo the last marker (box selections are undone first) or clear all
Dark purple area = currently selected region, bright purple outline = selection boundary, updated in real-time
During processing, a floating progress bar appears at the bottom (model download percentage / encoding / decoding), cursor changes to wait style; first-time click select requires ~40MB model download, please be patient
Recommended workflow: Box select the entire body first → then use "Exclude" mode to remove incorrectly selected background (hair gaps, beside arms)
Click Select mode starts from scratch, not refining the auto cutout result. Box select can quickly capture the complete subject — much more convenient than clicking many points.
Refine Mode (Enabled by Default, Recommended)
Workflow: First use "Auto Cutout" to get a result → switch to "Click Select / Refine" (refine mode is checked by default)
Exclude: Click on residual background areas (e.g., color patches at waistbands or hems that weren't fully removed), SAM will identify the object and subtract it from the result
Keep: Click on incorrectly removed areas (e.g., cutout holes, semi-transparent areas), SAM identifies and adds them back
Undo: Each refinement can be undone step by step (keeps last 8 steps); "Clear" restores to the pre-refinement result
Box refinements leave green rectangle markers, visible like point markers, and roll back together on undo
Dark purple overlay always shows the currently retained area — what you see is what you get
Refine mode = auto cutout as base + SAM local add/remove — much easier and better than starting from scratch with click select.
Batch Processing
Drag multiple images onto the page, or Ctrl+V to paste screenshots
Auto cutout after adding (on by default): images start processing immediately on drag/paste, no button needed; to adjust settings first, toggle off in the left panel
You can also click "Auto All" to batch process all pending images in the queue
When done, click "Download ZIP" to download all results at once
Background & Export
After cutout, choose background: Transparent / White / Red / Blue / Custom color
Gradient backgrounds: Built-in Twilight Purple / Warm Apricot / Mint gradients, great for poster backgrounds
Image background: Select "Image background (upload)…" to upload any landscape/scene as background, auto-fitted
Original background effects: Select "Original BG · Blur/Grayscale/Sepia" — subject stays sharp while background gets blurred or stylized
Remove white/gray edges: When enabled, bleeds foreground color into semi-transparent edges to eliminate white/gray halos around hair. Recommended when using dark backgrounds
Residue Cleanup / Subject Completion: If semi-transparent "ghost shadows" remain after cutout (e.g., sofa/table residue), drag slider right to remove; if subject appears semi-transparent (e.g., pants/skirts not fully cut), drag slider left to fill solid. Takes effect instantly, no re-cutout needed
Semi-transparency detection: If the status bar shows "~N% of area is semi-transparent" after auto cutout, there may be residue or incomplete cutout — follow the hint to use the slider or refine mode
Edge Feathering: Adjust slider (0-10px) to smooth cutout edges and reduce jagging. Default 2px, try 3-4px for hair detail
Layout Adjustment
Purple dividers between columns highlight on hover
Drag dividers left/right to adjust the three-column widths
Layout is auto-saved and restored on next visit
Cutout Quality Notes
Auto Mode (ISNet / RMBG): Best for photos with clear subjects, recommended as first choice. ISNet is fast but may lose subject details in complex scenes (e.g., seated legs); RMBG HD version has significantly better subject completeness — use it directly for complex images
Click Select Mode (SlimSAM): Good for specifying which object to cut out, but the lightweight model has limited performance on hair and complex edges
Small accessories (ribbon ends, pendants, etc.) may be missed by any auto model — use "Click Select / Refine" mode and click "Keep" to add them back
If click select results are poor, try switching to auto mode or using a higher-quality image
First use requires internet to download models (~40-80MB). After that, browser cache is used and works offline.
If you encounter a "Failed to fetch" error, the model download failed. Check your network connection or refresh the page to retry. Models are hosted on hf-mirror.com.
Troubleshooting
Auto cutout failed: The status bar shows the specific failure reason (network issue/image format/insufficient memory, etc.)
Network issues: Models are hosted on cdn.jsdelivr.net and hf-mirror.com — ensure both domains are accessible
Image formats: Supports JPG/PNG/WebP and other common formats; GIF animations are not supported
Insufficient memory: Very large images (>4000px) may cause memory issues — try compressing first
Batch failures: During batch processing, each image's failure reason is displayed for easy troubleshooting
HD RMBG is slow: First use requires downloading the 44MB model (via hf-mirror); with a dedicated GPU it takes ~10 seconds, CPU-only inference takes 1-2 minutes — this is normal, please be patient
About AI Background Removal
Free online AI background removal tool. Uses deep learning models to automatically remove image backgrounds. Supports precise cutout for people, objects, and animals with excellent edge detail handling. Runs entirely in your browser — images are never uploaded to any server, ensuring 100% privacy.
How to Use
Open the tool and drag & drop images or click to select files.
Choose a removal mode (Fast or Precise) and wait for the AI model to process.
Use the point-select repair feature to fine-tune edges if needed.
Select a background style (transparent, solid color, or custom), then download.
Frequently Asked Questions
How accurate is the AI background removal?
This tool uses advanced deep learning models like ISNET and RMBG to automatically identify foreground subjects and precisely remove backgrounds. It handles complex edges such as hair strands and semi-transparent objects well. Choose between a Fast model (40MB) for speed or a Precise model (80MB) for better quality.
What image formats are supported?
JPG, PNG, WebP, and BMP are supported. Output defaults to transparent PNG. You can also choose a solid color background before downloading. Individual files should be under 50MB.
What does the background look like after processing?
By default, images are output as transparent PNGs. You can also replace the background with solid white, blue, red, or any custom color, or upload a custom background image for compositing.
Can I process multiple images at once?
Yes, batch processing is fully supported. Select or drag multiple images at once and the tool will process them sequentially. Download results individually or as a ZIP package.
Is my image data safe?
Absolutely. All processing runs locally in your browser — the AI model runs on your device. Images are never uploaded to any server. All data is automatically cleared when you close the page.