In the room, not pasted in front of it.
Swapping the background behind a person is the easy version, and it always reads as a sticker. A body posed for a studio backdrop does not belong to a kitchen. Scene integration keeps who they are and re solves how they stand.
Lock the identity. Re solve everything the room owns.
Face, hair, body and wardrobe are preserved exactly. Pose, stance, orientation, gaze, scale, floor contact, occlusion and lighting are all re solved for the actual space, because those belonged to a different photograph and have no business surviving the move.
Six axes, reviewed before the footage is made.
Identity and scale
Still the same person, and the right size for the room. An identity miss is never auto corrected, because that is a human decision rather than something to spend money iterating on.
Grounding and occlusion
Feet meet the floor with real contact shadows, and foreground objects cross in front where they should.
Light and texture
Direction, softness and grain match the plate rather than the studio the actor was originally lit in.
The first frame decides the whole shot.
Whatever image the video model starts from is what it continues. Start from a studio portrait and the generated video spends its opening moments reconciling a room it was not given. Start from a plate that is already in the location and the footage opens where it should.
An honest limit
One composed plate serves one aspect ratio and one camera position, which is correct for a photograph and a real constraint for you to know about. Generating a second shape means composing a second plate. Fast composite mode, which preserves pose and swaps only the background, is still the right choice for quick work where the room is scenery rather than the point.
Put your cast somewhere real.
Start with the premium location library, or upload the actual room.