How to put yourself in a scene that does not exist

A generated background is the cheapest location there is: no travel, no permit, no weather, and it can be a place that does not exist. Combining it with a real person is also the oldest trick in television. The two halves are easy. The join is where everything goes wrong, and it goes wrong for reasons that were decided before either half was made.

Decide the shot before you shoot either half

The composite will only work if the person and the place were photographed, or appear to have been, by the same camera. That means three things agree: where the light comes from, how high the camera is, and how wide the lens is.

So decide those first, on paper. "Camera at chest height, slightly wide lens, key light from the upper left, soft, warm." Then shoot the person to that description and generate the place to the same description. Most failed composites skipped this step and tried to reconcile two shots that were never going to match: a person lit from the right standing in a room lit from the left, filmed from eye level in a room rendered from a drone.

Shoot for the key

The chroma keyer is forgiving, but the thing it cannot forgive is an uneven screen. Wrinkles, shadows and hot spots are all slightly different greens, and each one is a place where the key either eats the subject or leaves a fringe. Light the screen flat and separately from the person, stand the person well in front of it so their own shadow does not land on it, and the key takes seconds.

A few more things the keyer notices. Green reflects, so no shiny jewellery and no light-coloured clothing on the side nearest the screen. Hair is where keys fail, so hair against a well-lit screen with a little back light on it is the difference between a clean edge and a halo. And frame wider than you need: the composite will want room to move the person around.

Lock the camera. A person who was filmed handheld in front of a static background floats, because the background does not shake with them.

Generate the place to match

Generate the background as a still first, not as a video, and put the camera description in the prompt in plain words: "eye-level, 35mm, the floor visible in the lower third, soft window light from the left". A background with its floor visible is far easier to stand someone on than one that starts at knee height, and a lens that matches the one you shot with keeps the perspective lines agreeing.

Leave an empty space where the person will stand. A room with a chair in the middle of it is a room with a chair growing out of someone's hip. Prompt for the space: "the centre of the frame is empty floor".

Generate four and pick the one whose light most resembles the footage. Then, if the scene should move, animate the still with the gentlest motion that sells it: curtains, dust in a shaft of light, a slow push. The person is static relative to the camera, so a background that drifts sideways gives the game away; a background that pushes in slowly is fine, because the person can be scaled by the same amount with two keyframes.

Match the resolution. The background will fill the frame, so generate it at the highest resolution the model offers and upscale it to the export size. A soft background behind a sharp person is one of the quickest tells there is.

Build the composite

Two video lanes: the background on the lower, the keyed footage on the upper. Open the keyer on the upper clip, pick the green, raise similarity until the screen goes and back off before the edge erodes. A little edge blend softens the cut-out, and spill suppression takes the green cast off hair and shoulders. The preview shows the live composite over the real background, so tune it while looking at the join rather than at a grey checkerboard.

Then place the person. Scale and position are on the upper clip, and the thing to get right is the feet. If the floor is visible, the feet must sit on it, at a size consistent with where on the floor they are; a person too large for their position in the room is the error the eye catches first. If the shot is waist-up, frame the footage so the bottom of the frame cuts them, and the floor problem disappears.

The grade is the join

With the key clean and the placement right, the composite will still look like two images, and it is the grade that turns it into one.

Grade the person to the room, never the room to the person, because the room is the thing the viewer is reading as reality. Match the black level first: a person whose shadows are deeper than the room's shadows looks cut out regardless of anything else. Then the colour of the light: if the room is warm, warm the person. Then contrast, usually down a touch on the footage, because camera footage is crisper than a generated scene and crisp on soft reads as pasted.

Last, the small things that convince. A soft shadow under the person, which the keyed footage has lost: a dark, blurred, low-opacity duplicate of the keyed clip, squashed and placed at the feet, is a shadow. A faint touch of the room's colour on the person's edges. And a grain or a softness applied across both lanes together, so that the two images share a texture; nothing says "one image" like one grain over both.

When to let the background be a plate

Not every composite needs a moving scene. For a presenter, an explainer, a talking-head with a place behind it, a still background with a slight Ken Burns move, matched by the same gentle scale on the person, does everything a generated video would and costs a fraction of it. Save the moving background for a shot where the place has to do something: weather, a crowd, a passing train.

The order that works

1. Decide the shot on paper: camera height, lens, light direction, for both halves.
2. Shoot on a flat, evenly lit screen, person well in front, camera locked, hair back-lit.
3. Generate the place as a still to the same description, with its floor visible and the centre empty; animate it gently only if it must move.
4. Key, then place: background below, footage above, feet on the floor at the right size.
5. Grade the person to the room: blacks, light colour, contrast.
6. Add the shadow, share a grain, and judge the join at full size.