Ken Burns 3D Effect
StyleFrame Team8 min read
The 3D Ken Burns effect is the classic still-image pan and zoom with real parallax added, so foreground elements move faster than the background as a virtual camera travels through the scene. You build it by separating a photo into depth layers, or by generating a depth map, and then animating a 3D camera instead of scaling a flat image. In After Effects, Blender, Cinema 4D, or Nuke, the work comes down to three things: clean depth separation, believable fill-in for the areas the camera reveals, and a restrained camera move.
What Makes It "3D" Compared to the Classic Effect
The traditional Ken Burns effect is a keyframed scale and position change on a single flat image. In Premiere Pro that means two keyframes on Scale and Position, with the second placed at the end of the clip so the move doesn't stall. It works for documentary stills, but it reads as a crop sliding across a photograph, because every pixel moves at the same rate.
The 3D version fixes this with motion parallax. Real cameras see near objects shift more than distant ones, and the brain uses that cue to read depth. Once the foreground, midground, and background travel at different speeds, the viewpoint seems to enter the picture instead of scanning across it.
The cost is that the camera now sees things the photo never captured: the area behind a person's head, the edge of a column, the wall behind a lamp post. Most of the craft in this effect is hiding or reconstructing that missing information.
Method 1: Layered 2.5D in After Effects
This is the most controllable approach and a common first choice for motion designers.
Step 1: Split the image into depth planes
Cut the photo into layers in Photoshop or directly in After Effects. A typical split is foreground subject, one or two midground elements, a background, and sometimes a sky. Use as few layers as the move requires. Every extra cutout is another edge to clean up.
Step 2: Rebuild what's behind each cutout
Each layer needs painted or cloned pixels behind the object that was removed. Extend the plate beyond the cutout by more than you think you need, since the camera will reveal it. Content-aware fill, clone work, or generative fill can do the first pass, but check edges and texture continuity by hand.
Step 3: Place layers in Z space and add a camera
Turn on the 3D switch for every layer and arrange them along the Z axis. Create a camera and animate it, usually with a slow push-in plus a small lateral move. Keep the focal length consistent with the apparent lens of the original photo. A mismatch makes perspective feel wrong even when the parallax is technically correct.
Scale each layer after moving it back so it still fills the same area of frame. If you push a layer 500 units away, compensate with scale. Otherwise the composition shrinks.
Step 4: Finish the move
- Ease the camera with the graph editor. Linear moves look mechanical.
- Add depth of field using the camera's focus distance and aperture, tied to the same Z positions.
- Add subtle atmospheric haze to distant layers.
- Apply motion blur and a consistent grain pass over everything so the layers sit together.
Method 2: Depth-Map Displacement
If you don't want to hand-cut layers, generate a depth map and use it to displace geometry. A monocular depth estimator produces a grayscale image where brightness stands for distance. You then project the photo onto a subdivided plane and displace it by that map.
- Blender: Add a subdivided plane, assign the photo as an emission or unlit texture, and drive a Displace modifier with the depth map. Animate a camera over it.
- Cinema 4D: Use a displacer deformer or a displacement channel on a high-resolution plane, with the image applied through a camera-projected or UV-mapped material.
- After Effects: Use the depth map as a luma matte or displacement source with 3D layers, or bring in geometry from a 3D application.
- Nuke: Project the image through a Project3D node onto cards or displaced geometry, then render through a ScanlineRender node with a moving camera. This suits shots that need to sit inside a larger comp.
Displacement is fast, but it has predictable weaknesses. Depth maps from a single image often blur at object boundaries, so edges stretch like rubber when the camera moves. Thin structures such as hair, wires, and railings tend to smear. Keep the camera move small, or combine approaches: use displacement for the environment and separate layers for key subjects.
Method 3: The Research Approach, Single-Image Synthesis
The phrase "3D Ken Burns effect" is also tied to a 2019 paper in ACM Transactions on Graphics by Simon Niklaus, Long Mai, Jimei Yang, and Feng Liu. Their system generates the effect from a single image. It estimates depth with a semantic-aware neural network, adjusts that depth using segmentation, and refines it at object boundaries. It then converts the image into a point cloud and renders frames from a virtual camera path. To handle disocclusions, it inpaints both color and depth for the extreme camera positions.
It offers a fully automatic mode and an interactive mode where you control the camera. A PyTorch reference implementation is public on GitHub. It relies on CUDA through CuPy, and video output needs moviepy. If you plan to use any of it commercially, read the licenses carefully. The repository is listed under "Other," and the accompanying dataset is licensed for non-commercial use only.
For working artists, the paper is useful mainly because it names the problems you hit anyway: inaccurate depth edges, geometric distortion, and missing information at the extremes of the camera path. It also describes a workable pipeline of depth, point cloud, camera render, and inpaint. You can mirror that pipeline by hand with tools you already own.
Where Styleframe Fits in This Workflow
Styleframe is an AI video and motion tool. Tools like it can help with some of the tedious prep in a 3D Ken Burns build, such as isolating subjects, upscaling small archival scans, and matching styles across stills, while compositing stays in your hands. A push-in crops into the source and exposes every missing pixel, so source resolution matters. For artists who build parallax moves regularly, the practical gain is in the prep work: cutouts, fill-in, upscaling, and style matching. The final composite stays yours.
Camera Moves That Hold Up
The 3D Ken Burns effect rarely survives an ambitious move. These guidelines keep the illusion intact:
- Move less than you want to. The more the camera translates, the more disocclusion it exposes. A slow push-in with a small lateral drift covers most documentary and title-sequence uses.
- Favor dolly over pan. A camera that translates produces parallax. One that only rotates in place doesn't.
- Start from the composition. Choose the end frame first. Then frame the start so the move lands on the subject you want the viewer to notice.
- Respect the source's lens. Wide-angle photos tolerate forward moves well. Telephoto compression gives you little depth separation, so exaggerating the parallax looks wrong.
- Keep the timing consistent with the edit. A move that fits a six-second hold shouldn't be stretched to fill ten. Slow it down only if the extra length doesn't expose more of the fill.
Common Problems and Fixes
Stretched or rubbery edges. Usually caused by a soft depth map or mesh density that's too low. Sharpen the depth edges, add polygons, or separate the subject into its own layer.
Visible seams between layers. Feather and defringe your mattes, extend the background behind them, and add a unifying grain and subtle blur pass at the end.
Flat-looking depth. The layers are too close together, or the scale compensation is off. Increase Z separation and re-check each layer's size.
Soft output on push-ins. Source stills are often small. Upscale before the camera move, not after, and check the final frame at 100%.
Stylistic mismatch between layers. If you're mixing scans, painted fills, and generated elements, unify them with a style pass or consistent color grade before the camera move.
Choosing a Method
- Use layered 2.5D in After Effects when you need clean control and expect client revisions.
- Use depth-map displacement in Blender, Cinema 4D, or Nuke for environments, architectural shots, or when the still has to integrate with 3D elements.
- Use single-image synthesis research as a reference for how depth, point clouds, and inpainting fit together, or as a starting point if you have the engineering support to adapt it.
- Combine them when a shot is hard. A displaced environment with hand-cut hero layers is often the most reliable result.
Whatever the method, the principle is the same: the camera is only convincing if the picture behind it holds up. Spend your time on depth separation, fill-in, and edge quality, and keep the camera move modest.