1 The idea
Move the whole picture with a Transform and everything shifts by the same amount, like one flat photo. A real camera sliding sideways makes near things race and far things crawl. That difference is parallax, and it is what tells the eye a scene has depth.
2.5D means cutting the plate into layers (sky, hills, mid ground, near ground), putting each on a card at the right distance and filming them with a 3D camera. A farther card is scaled up with its distance, so the still frame looks identical. The only difference shows when the camera moves.
shift (px) = f_px · T / Z · f_px = 320 / tan(½ hFOV) ≈ 6222 Hands-on
Switch to 2.5D, then set the near, mid and hills card distances so each coloured dot sits on its orange ring (within 8%)
Watch the strip under the picture: each orange ring marks the true depth of that point
3 Deep dive
Why near things move faster
A pinhole camera puts a point at depth Z on screen at f·X/Z. Slide the camera sideways by T and that point moves f·T/Z on screen. Z sits under the fraction, so deeper things always move less.
Our plate camera (35 mm on a 36 mm sensor) has f ≈ 622 px across a 640-wide frame. Slide 1 m and a rock 10 m deep moves 62 px, while a hill at about 89 m moves only about 7 px.
| Depth | Shift per 1 m | In our plate |
|---|---|---|
| 7 m | 89 px | near rock, bottom left |
| 18 m | 35 px | mid-left rock |
| 89 m | 7 px | front hill |
| ∞ | 0 px | sky |
Farther cards must grow with distance
Push a card twice as far without scaling it and it looks half the size in camera. Scale it with distance (size ∝ distance) so it fills the same frame, and the still image matches the plate pixel for pixel.
This rule means card distance “costs nothing” while the camera is still. Distance only shows once the camera changes position, so always check with a sideways or forward move.
- 1Split layers by depth or roto
- 2Premultiply each layer
- 3Card at true depth + scale
- 4Move the camera to check
Layer edges and overscan
A hard-cut layer edge shows as a line once the camera moves. In the figure each near layer fades out on its far side, and the layer behind is fully solid under that fade, so the still frame has no dark rims or seams.
When a near card moves a lot, its frame edge slides into view. You need picture past the edge (overscan). The figure just repeats the edge pixels, which only holds up for a small move or truly featureless texture like this sand. For a bigger move, or anything with clear detail, paint real overscan (see P5).
Let the depth decide where a card goes, not a hunch
When cards are not enough
A card has only one depth. Ground running from 5 m to 40 m on one card slides as a single sheet. With a big move its near and far edges look wrong. Split it into more layers, or use geometry that follows the real ground and project onto it (next lesson).
4 In real work
In Nuke
- Cut each layer with a Roto or a depth key, then Premult it
- Feed each layer into its own Card, place it at its true depth and scale it with that distance
- Gather the cards in a Scene and render through ScanlineRender with the tracked or hand-animated Camera
- Check that a still frame matches the plate, then play the shot to judge the parallax
Common mistakes → the fix
- ✗ Pushing a card back without scaling it, then fixing the size with a 2D Transform✓ Scale the card with its distance in 3D; the still frame then matches and the parallax is right
- ✗ Cutting layer edges hard✓ Keep edges soft and make sure the layer behind is solid under them
- ✗ Judging card depth on a still frame✓ Depth only shows when the camera moves: play or scrub every time
- ✗ No picture past the frame edge of near cards✓ Paint overscan far enough for the camera move; stretching the edge pixels only works for small moves on featureless texture (see P5)
Before you send the shot
- A still frame matches the original plate
- Cards are ordered near to far by real depth
- No dark rims or seams between layers while the camera moves
- Near cards’ frame edges never show during the shot
5 Check yourself
The camera slides 1 m. A rock 10 m deep shifts 62 px. About how far does a hill 100 m deep shift?
🔒 Answer the question above and this card opens
1. You move a card from 10 m to 20 m. What must happen to its size for the still frame to stay the same?
2. The camera slides 2 m. Roughly how far does a card at 20 m move on screen (f ≈ 622 px)?
Where the numbers come from: The scene is rebuilt in your browser from our desert plate and its Blender depth pass (35 mm lens, 36 mm sensor, 1.6 m high, tilted 4° down). Parallax numbers use the pinhole formula: move the camera T m and a point Z m deep shifts ≈ 622·T/Z px in a 640-wide frame. The layer split, the “stretch” measure, the depth mesh and the automatic clean plate are teaching versions (object edges and rock removal are simplified), not Nuke output.