This workflow extracts characters or objects from a scene, and creates variations of background-free images based on the selected area. The generated images usually (small chance of failure) have transparent background, allowing you to paste them into new backgrounds using image editing tools.

The workflow uses IPAdapter to approximately extract selected region, and the composition and style should be mostly retained, but the new images are not exact copies, and precise identity of characters may not be retained (if that is needed, an additional face swapping step would be helpful).

The initial masking can be done via manual masking or SAM. For the final image, after pasting into a new background, some additional post-processing such as low-strength image-to-image denoising or relighting might be helpful for improving quality.