Fast Vehicle and Chase Shots in AI Video: How to Plan Speed Before the Camera Gets Lost

Introduction

The chase in your head is clear: a high shot over the highway, a dive between the towers, the car screaming past in profile, a pull back to the whole city. What comes back is four clips that are each fast and each somewhere else. In one the car goes left, in the next it goes right, in the third the camera is suddenly on the ground, and the cut looks like a highlight reel of four different chases.

Fast shots are unforgiving because speed multiplies camera travel. A slow walk gives the camera a meter to get wrong; a car at speed gives it fifty. The only way to keep a chase readable is to plan where the camera is at every position before anything is generated, and to write direction and speed into each shot so the model has nothing to guess. This guide shows how to plan fast vehicle and chase shots for AI video, position by position, and cut them so the speed carries across.

Why speed breaks the camera

Three things happen at speed that do not happen in a walking shot:

• Direction becomes the whole shot. If the car travels left to right in one clip and right to left in the next, the audience reads a U-turn. Nothing else about the clips matters.
• The camera has to match speed or fall behind. A tracking shot that drifts slower than the car turns into a shot of the car leaving. The speed relationship must be stated.
• Every second covers more space. The background changes fast, so continuity of the environment — the same towers, the same road — has to come from a location asset, not from luck.

None of these are solved by longer prompts. They are solved by splitting the chase into positions and writing three facts into each: direction, speed relationship, and end point.

Rules for fast shots

• One camera position per shot. High angle is a shot. Dive is a shot. Side is a shot. Do not ask one generation to do two.
• State screen direction in every shot. "Left to right" or "away from camera." Repeat it even when it feels obvious; it is the first thing that flips.
• State the speed relationship. "Camera matches the car's speed," "car pulls ahead of camera," "camera is static and the car passes." One of the three, every time.
• Give every shot an end point. "Ends as the car clears the gap." Without an end point the model decides when the move is over, and it is never where you want the cut.
• Keep the environment as an asset. The road, the towers, the tunnel: a location asset referenced with @, so the same city appears in all four positions.
• Shorter is faster. A three-second clip of a car at speed reads faster than a ten-second one, because the cut does the work. Plan short.

For how the shot description carries camera language, see the camera control guide. For drawing the positions before you write them, the 3D previz guide has the workflow; a chase is exactly the case where it pays off.

How to plan a chase by camera position, step by step

1. Write the route in one line. Where the vehicle starts, what it passes, where it ends. "Enters the expressway from the ramp, threads the two towers, exits toward the bridge."
2. List the positions. Usually four to six for a short chase: an establishing high angle, one or two close tracking positions, one static pass-by, one wide to close. Name them A, B, C, D.
3. Assign a direction to the whole chase. Left to right, or toward camera. Every position keeps it unless a cut is meant to reverse it on purpose.
4. Write one shot card per position. Reference @vehicle and @location. Each card: position, direction, speed relationship, one action, end point. Nothing about the car's paint or the driver's face; the assets carry that.
5. Generate the stills first and line them up in order. Check that the direction reads the same left to right across all of them. Fix any card that flips before generating video.
6. Generate video for the positions that passed. Batch through the AI Chat Panel if the cards are ready: "Generate videos for Shots 1, 2, and 3."
7. Cut in Edit, short. Trim each clip to the part where the speed is clearest — often the middle second and a half. Put the wide last so the sequence exhales.
8. Review the cut for direction and speed only. Paint, lighting, reflections can vary between clips; the audience will not notice at speed. A flipped direction they will notice instantly.

Example 1: Position A, the high angle

High-angle establishing shot over @Neon Expressway at night, camera slowly descending toward the road, no pan. @Vela's car enters from the bottom of frame, small, headlights on, and drives away from camera up the center lane at speed. Two lanes, towers on both sides, the gap between the two tallest towers directly ahead. Ends as the car reaches the gap. Rain-wet asphalt, sign reflections.

Example 2: Position C, the side pass

Static camera at road level between the two towers of @Neon Expressway, facing across the lanes. @Vela's car passes left to right at full speed, filling the frame for a moment, neon reflections streaking across the door. The camera does not pan or follow; the car enters, passes, and exits frame right. Ends one beat after the car is gone, with the empty wet road. Under one second of car in frame.

Example 3: Position D, the pull-back wide

Camera behind and above @Vela's car, matching its speed as it exits the towers, then pulling up and back so the car shrinks toward the bridge and the whole @Neon Expressway skyline opens into frame. Direction stays left to right. Camera rises smoothly, no shake. Ends on the wide with the car a point of light near the bridge. Night, city glow on low cloud.

The most common ways chase sequences break

• Direction flips on a cut. One card said "left to right," the next said nothing. State direction in every card.
• The camera falls behind. The tracking shot became a shot of taillights. Add "camera matches the car's speed."
• The environment changed. New towers in every clip. Reference the location asset with @ in all of them.
• Two moves in one shot. The dive and the pull-back in one generation; one of them vanished. Split.
• Clips too long. Ten seconds of a car at speed turns into a driving video. Trim to the fastest second and a half.

FAQ

How many positions does a short chase need? Four is enough for a ten-second sequence: establish, close, pass, wide. Add positions only when a beat has nowhere to land.

Does the vehicle need to be an asset? If it appears in more than one shot, yes. It is the same identity problem as a character: build it once, reference it with @, and keep its description out of the shot text.

Can the camera be handheld for a chase? It can, but say so explicitly and keep it to one position. Handheld everywhere reads as noise, not speed.

What about a chase on foot? Same rules, smaller numbers. Direction, speed relationship, end point, one position per shot. The runner is the vehicle.

Next step

Take the chase you have been re-rolling and write it as four positions on one page: establish, close, pass, wide, each with direction, speed relationship, and end point. Build the vehicle and the location as assets if they are not already. Open the project in ArcLoop, generate the four stills in order, and check the direction reads before any video.