Open Background Remover
Import an image and choose fast colour removal or AI background removal.
ImgIng / Remove image backgrounds with AI running on your device
LOCAL AI · BACKGROUND REMOVALRemove image backgrounds with AI running on your device
Use connected-edge colour removal for simple studio backgrounds, or download one of three AI models for complex people, hair and translucent material. Source pixels never leave the device.
Local WebGPU / WASMThree AI model tiersTrue 1 px brushBackground and layer layout
Capability quick facts
These facts describe the current product, not unshipped roadmap work.
- Processing
- Connected-colour removal and all three AI tiers run locally
- Models
- ISNet INT8 / BEN2 FP16 / BiRefNet HR-Matting FP16
- Refinement
- True 1 px brush, undo/redo and black/white edge inspection
- Outputs
- Transparent WebP / PNG or local multi-background composition
Reviewed by the DataDance product and engineering team · Published · Updated
Finish in three steps
Confirm important settings before export; the processing location is always disclosed.
Choose a model and run
Fast AI is about 42 MB, Pro AI about 219 MB, and Ultimate AI about 447 MB.
Refine, compose and export
Inspect the edge, refine with a 1 px brush, then set a colour or image background and arrange the subject.
Fast colour removal or AI?
Connected-edge removal is ideal for product art and graphics surrounded by a consistent colour. It removes only similar pixels connected to the canvas edge, protecting matching details inside the subject. People, hair and complex scenes are better suited to AI.
Why does the default preserve the model edge?
The default path uses the model alpha channel pixel for pixel. Edge contraction, feathering and colour decontamination run only when you enable them, so compact preview, professional editor and full-size export can share the same mask.
How can I compose the cutout without another editor?
Professional refinement can keep transparency or use a solid colour or local image as the background. Move and scale the subject and background independently, then choose the source canvas, 1:1, 4:5, 16:9 or a custom size. Dragging updates CSS transforms only; full-resolution pixel compositing waits until export to keep large-image interaction responsive.
Model and background files stay in local browser memory. Downloading a model does not upload the source image.
Updated 2026-08-20 · Live capability detection inside the tool is authoritative
Frequently asked questions
These visible answers match the current product behaviour and structured data.
Which AI model is best for a modest computer?
ISNet INT8 Fast AI is recommended by default. Its smaller download and faster inference suit CPU-only and lower-spec devices.
Does the result preserve partial transparency?
Yes. The mask retains the complete 0–255 alpha range, and alpha-capable outputs such as WebP and PNG preserve soft edges.
Why might the model download again?
The browser normally reuses Cache Storage. Clearing site data, using a private window or browser cache eviction can require another download.
Explore ImgIng
Each task page documents real settings, limits and format advice—not keyword-swapped duplicates.