Object remover

No. LaMa runs in a worker in your browser. Only public model files are downloaded from Hugging Face; your photo and selection stay on your device.

The browser downloads a 107 MB model for compatible GPUs or a 62 MB CPU model. Downloads are cached when browser storage permits. If GPU processing fails, the CPU model may also need to be downloaded.

The AI invents a plausible replacement from nearby pixels. Small objects against grass, walls, or other textures are usually easier than large objects or complex patterns. Cover the entire object and its shadow; undo an edit and try a different selection if needed.

No. It processes a crop around your selection at 512 × 512 model resolution, then blends the painted area back into the image at its original size. Pixels outside the selection and the original transparency are preserved. Large selections can have less detail.

PNG, JPG, WebP, and AVIF files up to 20 MiB and 16 megapixels, with neither side over 8192 pixels. AVIF requires browser decoding support. Downloads are PNG still images; metadata is not retained.

Yes. Focus the image editor, move the brush with arrow keys, and press Space or Enter to paint. Hold Shift with an arrow key to move one image pixel. Brush size, undo, clear selection, and removal controls are also keyboard accessible.

No. Object removal fills an area you paint over while keeping the rest of the scene. Background removal automatically separates the subject and exports a transparent cutout.

No account, API key or paid credits are needed. The model runs on your device after its files download. Available memory, browser support and image limits still apply.