Check foreground selection, existing transparency and source contrast when automatic removal leaves almost no visible subject.
Compare with the original preview
First confirm that the correct image opened and that its subject was visible before processing. A mostly transparent source, a corrupted export or a subject outside the chosen crop can cause confusion before AI selection even begins. If the original looks right but the result is nearly empty, the model may not have identified a clear foreground.
Choose stronger visual separation
Try a photograph with a distinct main subject, clearer boundaries and less competing clutter. A landscape, patterned surface or abstract texture may not contain a single object that this foreground model can usefully isolate. The tool is not a general command-based selector and cannot be told to keep a specific tiny region by name.
For example, a small dark ornament on a crowded dark shelf may disappear, while a closer photograph of the same ornament against a plain wall gives a much clearer input. Use a prepared crop only if it retains all necessary edges.
Do not build on an empty result
Adding a colored background to a nearly empty mask does not fix subject selection. Return to the original rather than reprocessing the blank cutout. If the same suitable photograph repeatedly fails, use a manual masking tool. Downloading a file successfully proves only that an image was exported; inspect that the intended subject actually remains before treating it as usable.