An image mask is a control map that hides, reveals, or partially shows areas of an image. In a typical grayscale mask, white reveals, black hides, and gray creates partial visibility. The exact interface varies by editor, but the important property is reversibility: the underlying image can remain intact while the visible result changes.
Masks are used for background removal, local color corrections, compositing, selective blur, and many AI-assisted selections. They solve a different problem from cropping, which changes the canvas boundary, and from erasing, which can permanently remove pixel data.
How a mask controls visibility
A mask can be understood as a second image aligned with the first. Each position says how visible the corresponding source pixel should be. Hard black-and-white transitions create a cut-paper edge; intermediate values create a soft transition. Editors often display the mask separately so you can paint or inspect that visibility map directly.
This control is separate from the source color. You can correct the boundary without repainting the subject, or reuse the same subject against a new background. When an export format supports alpha, the visible result of the mask may become per-pixel transparency in the delivered file.
Selection, segmentation, and matting
A selection identifies an area for an operation. Segmentation predicts regions or objects. A mask is the editable representation that can hold or refine the result. Matting goes further at difficult boundaries by estimating partial foreground coverage, which is important for hair, fur, glass, smoke, and motion blur.
Promptable segmentation research shows how a model can propose masks from points, boxes, or other prompts, but a predicted region is not automatically a finished production edge. Review the output at the intended display size and on more than one background before accepting it.
A practical non-destructive workflow
Duplicate or preserve the source layer, create the initial selection, convert it to a mask, and refine the mask rather than deleting pixels. Use a small brush at high zoom for local errors, but periodically return to normal viewing size. Overworking a boundary at extreme zoom can make it unnaturally crisp when seen in context.
Name important masks and keep the editable project when the asset may be reused. Export a transparent delivery file only after the mask has been checked on light and dark backgrounds. This separates the master asset from its channel-specific outputs and makes later corrections much safer.
Common mask failures to check
Look for missing interior holes, clipped thin parts, leftover background islands, color spill, and an edge that is either uniformly hard or uniformly soft. A strong mask follows the material: a metal rim may need a crisp boundary, while hair may need partial coverage. One global feather value rarely fits every edge.
If the subject and background are very similar, the source may not contain enough contrast for a reliable automatic separation. Improve the selection with a tighter prompt or manual input; if important detail was never captured, reshooting can be more honest and efficient than inventing a boundary.
Conclusion
A mask is best treated as editable visibility, not as deleted content. Preserve the source, refine different materials selectively, and validate the output in context. Continue with background-removal edge troubleshooting or learn what an alpha channel stores.
Editorial note: This explainer was drafted with AI assistance, then checked and edited against the sources listed below on September 13, 2026. It does not report an unperformed product test.
Sources
Explore the complete topic guide
Improve this guide
Report a problem or suggest the next question
Submissions enter a private editorial queue. They are reviewed and deduplicated before any public update or new page is created.