You draw the boundary, the model repaints inside it. Everything outside the mask is preserved exactly — not re-rendered, not re-graded — which is what makes this the tool for surgical work: remove a bystander, change a sign, fix a hand, swap a colour on one object.
The mask is most of the craft. Too tight and the new content has nowhere to blend, leaving a visible seam. Too loose and you have handed the model territory you wanted kept. Feathered edges and a little breathing room around the subject produce joins that survive scrutiny.
The model also reads the surrounding pixels for context, so it matches light and perspective from what is nearby. When it gets that wrong, the fill is competent and lit from the wrong direction, which reads as fake faster than a blurry patch would.
In practice
- Mask generously around the object, not tightly along its edge.
- The prompt describes what should be inside the mask, not the whole picture.
- Iterate on the same mask at different strengths rather than re-masking each time.
The mistake to avoid
Masking the object you want gone and prompting for the object. Prompt the replacement — an empty wall, a clear pavement — not the thing you are removing.
Where you will run into it
- Remove the Background From a Photo — Same job, different pipeline than video.
Related terms
Outpainting
Outpainting extends an image beyond its original borders, generating new content that continues the scene outward.
Image-to-image
Image-to-image takes a picture as its primary input and returns a changed picture — restyled, corrected or varied — instead of inventing one from nothing.
Denoising strength
Denoising strength decides how much of your input picture gets thrown away before regeneration — low keeps it nearly intact, high keeps only the general shape.
Segmentation
Segmentation labels which pixels in each frame belong to a chosen object, producing a mask that other tools then act on.
Temporal consistency
Temporal consistency is how well a generated clip keeps things the same from one frame to the next — a shirt that stays the same colour, a background that stays put.
The all-in-one AI studio for creators. 60+ models for video, image, voice, music and lipsync in a single app.