1 The idea
Long lenses seem to "compress" the background, but perspective really comes from where the camera stands. From the same spot, a longer lens just crops. Back the camera off to keep the subject the same size, and only then does the background grow.
When putting CG in a plate, the camera distance has to be right, not just the focal length. With the wrong distance the CG perspective won’t sit in the plate even if the size looks right.
2 Hands-on
Keep "lock robot size" on and move the focal length
3 Deep dive
Relative size comes from distance
Things look smaller the further they are from the camera, so how big the hills look next to the robot depends on the ratio of their distances, not on the focal length.
Example: robot at 3 m, hills at 300 m: the hills are scaled down 100× relative to the robot. Back the camera off so the robot is at 12 m (hills at 309 m) and the ratio drops to about 26×: the hills look about 4× bigger next to the robot.
| Robot distance | Focal for same robot size | Hills vs robot |
|---|---|---|
| 3 m | 35 mm | 1× |
| 6 m | 70 mm | ≈ 2× |
| 12 m | 140 mm | ≈ 3.9× |
Do wide lenses really "distort faces"?
Faces look big-nosed on a wide lens because you have to stand very close to fill the frame, so the nose is much closer than the ears. Crop the centre of a wide shot from the same spot and the perspective is identical to a long lens (only resolution, depth of field and lens distortion differ).
Perspective = where the camera stands. Focal length = how much you crop.
How a dolly zoom works
The robot’s size in frame follows focal ÷ distance. To keep it the same while the camera moves, the focal must follow the distance exactly: focal = start focal × distance ÷ start distance. The background then stretches or shrinks around a robot that holds still.
- 1Pick the robot size
- 2Move the camera
- 3focal = f0 · d / d0
- 4Background grows/shrinks
Matching CG: get the distance right first
If the CG camera is too close and you fix the size with a wider focal, the robot reads the right size but its perspective is exaggerated: its lines converge more than the plate’s, and parallax goes wrong once the camera moves. Too far with a long lens looks too flat.
On set, measure the distance from camera to subject or to a floor mark, and the camera height, for example the distance to the rocks in the desert plate. That number sets the track’s scale correctly.
4 In real work
In Nuke
- Use the Camera from CameraTracker (may need NukeX) or from matchmove; don’t change the focal to resize the CG
- Wrong size: adjust the object or scene scale (scaling the whole scene with the camera doesn’t change the image), not the focal
- Dolly zoom: robot at the origin, camera on the z axis, focal expression such as 35 * translate.z / 3
- Check with a real-size Card or Cube, ScanlineRender and Merge over the plate; watch converging lines and parallax across the shot
Common mistakes → the fix
- ✗ Fixing CG that looks too big by changing the focal length✓ The focal changes the whole view; fix the scale or the distance
- ✗ Believing a long lens “compresses” the background by itself✓ The compression comes from standing far away; the long lens just crops
- ✗ Animating the focal linearly while the camera moves unevenly: the robot pulses in size✓ Link the focal to the distance with an expression
- ✗ No camera height or distance measured on set✓ Measure the distance to the subject or a marker and the camera height for every setup
Before you send the shot
- The CG camera position comes from the track, not a guess
- Scene scale is set from a real measurement
- Converging lines on the CG agree with the plate
- Parallax between CG and background holds across the shot
5 Check yourself
You want the hills behind bigger while the robot stays the same size. How?
🔒 Answer the question above and this card opens
1. From the same spot you shoot at 24 mm and 96 mm. Crop the centre quarter of the 24 mm width and scale it up. Compared with the 96 mm shot?
2. Dolly zoom: the camera backs off from 4 m to 8 m from the robot, starting at 30 mm. Which focal keeps the robot the same size?
Where the numbers come from: Lens maths is the standard thin-lens model (FOV, CoC, hyperfocal), using CoC = sensor diagonal ÷ 1500, a common convention rather than a fixed rule. Distortion uses the Brown–Conrady model (k1, k2). The real DOF images are Blender Cycles renders.