The original bullet-time shots were made with a ring of stills cameras fired in sequence, which is why the effect stayed out of reach for one person with one lens. The shot below was made from a single generated photograph in CapCut, at 16:9, in two requests. This article covers what has to be written into the prompt to keep the subject frozen, how to read how far the camera actually traveled, and what changes when the clip is ten seconds instead of five.
Ten seconds, one continuous circle around a man suspended in mid-air. Shown here as frames in sequence; the full video and a GIF version are supplied alongside this document.
Bullet time is three things holding still and one thing moving
Three things in a shot like this could move: the subject, the world around him, and the camera. The effect exists because only the last one does. That sounds obvious written down, and it is where the work goes. A photograph of a man in mid-air contains an unfinished jump, and the section below shows what came back when the prompt did not say so. Everything here follows from making the freeze explicit and then checking whether it held.
The man, the street and both clips here were generated, so nothing was filmed and nobody was photographed. If you want to build this from footage of a real person jumping, that is a different route and not the one described here. Sound, music and color grading are also outside this article.
Freeze a pose the body could not hold
A person standing still, photographed from a moving camera, reads as a person standing still. What makes an orbit read as stopped time is a pose that gravity would not allow to last: both feet off the ground, a body mid-rotation, and small matter suspended where it should be falling. The starting still was written for those three things, and it serves as the visual starting point for the generated video.
Photoreal wide shot, 16:9 landscape. A young man is frozen in mid-air above a wet city street at night, both feet off the ground, his body twisted in a mid-jump dance freeze with one arm stretched out and one knee tucked up. Water droplets and small specks of dust hang motionless in the air all around him. Neon shop signs and street lights sit out of focus behind him, and the wet asphalt reflects them. Low camera angle close to the ground, the man in the middle of the frame with open space on both sides of him. Plain black jacket and black trousers with no logos and no text. Sharp focus on the man, shallow depth of field.
The starting still, made in Image mode. The suspended droplets are doing as much work as the pose.
Two details in that prompt matter later. The man sits in the middle of the frame with space on both sides, because the camera is about to swing through that space. And the low angle close to the ground gives the orbit something to sweep past. The still was generated with the CapCut AI image generator in Video Studio's Image mode. This session used Seedream 4.5 at 16:9 and 2K, and the result measured 2560 by 1440. Before generating, check the current model options, resolution and credit estimate shown in your version.
Say what stays still, or the jump finishes
The still then goes back into the composer as an attachment, with the mode menu switched to Video, which is the CapCut AI video generator path in the same workspace. The settings chip opens a panel holding Video clip and Full video, then Model and quality, an Aspect ratio row and a Duration row. This run used Video clip, Seedance 2.0 Mini, 720p, 16:9, and 5s to begin with.
The panel behind the composer chip, here set for the ten-second version. The chip above it shows what will be sent.
Two clips were then generated from the same still at the same settings, differing only in what the prompt said about movement. The first named the camera move and left the subject alone. The second named the camera move and then spent three sentences saying that nothing else moves.
The camera moves in a smooth circular arc around the man, travelling to the right and keeping him in the middle of the frame the whole way, so the neon street behind him sweeps past. He is frozen in place: his pose does not change, his arms and legs do not move, his jacket does not move. The water droplets and specks hang still in the air. Nothing in the scene moves except the camera.
Same still, same settings, five seconds each. The only difference is the three sentences about what stays still.
The clip without those sentences orbited correctly and let the jump play out: by the five-second mark the man had come down and was crouching in the puddle. The camera move was not the hard part in either run. Reading the pair as one session's evidence, the freeze is the thing the prompt has to insist on, and the camera arc is the thing it can ask for once.
Use the subject as the protractor
Asking for a circle does not tell you how much circle you got. There is a way to read it off the picture without measuring anything: watch which side of the subject faces the lens. Chest at the start and back at the end is half a turn. Chest, back, chest is a full one. The background helps as a second opinion, since a landmark that leaves frame on one side and returns on the other has come all the way round.
The five-second version. His chest faces the lens at the start and his back faces it at the end, so the camera covered about half a turn.
The pose held for the whole five seconds. What the shot did not deliver was the promise in its own name, because half a circle is not a bullet-time 360.
Compare five and ten seconds for a full circle
The shot at the top of this article is the same still at 10s, with the camera sentence rewritten to ask for one continuous circle, a full 360 degrees, moving to the right and ending back where it started. The three sentences about the freeze were left as they were. Reading it with the same protractor: his chest faces the lens at 0.0 seconds, his back faces it at 4.9, and his chest faces it again at 9.8. The red storefront that fills the left of the opening frame is back in the closing one.
The ten-second version, sampled every 1.2 seconds. Chest, back, chest: one full turn, ending on the storefront it started on.
Before each request, the interface displayed an estimated credit cost for the selected model, duration, aspect ratio and resolution. The ten-second request used more credits than the five-second request in this session, but credit use can vary by model, settings, account and region. Check the current estimate shown in your version instead of carrying these session figures into a new project.
The settings line and the price line for the ten-second request, shown before it runs.
What the freeze did not hold
Two things drifted, and both are worth checking in your own results. In the five-second clip the curtain of water droplets that fills the right of the opening frame is mostly gone by the closing one, so the suspended matter faded even though the body did not. And in the ten-second clip a crown of splashed water appears in the puddle directly under him from about 2.5 seconds onward and then stays there, frozen, for the rest of the shot. It is not in the starting still. Nothing in the prompt asked for it.
The splash is a reasonable thing for the picture to contain, since a man in the air over a wet street has to have come from somewhere, and it holds still like everything else. It is still an invention, which is the point: the freeze applies to what the model chooses to draw, not only to what you gave it.
Give the circle a stop
A steady orbit at one speed is a camera move. What makes it read as bullet time is the rhythm around it: fast in, hold, fast out. That part happens after the clip exists, in the web editor. In the interface used for this build, selecting the timeline clip opened Speed and Curve controls in the right-hand panel. The available choices included Custom and presets such as Montage, Hero, Bullet, Jump Cut, Flash In and Flash Out; preset names and availability can vary by version or region.
The Curve tab, with the preset named Bullet applied. The graph is editable, and the panel reports what the change does to the clip's length.
Applying Bullet to the ten-second orbit turned it into 3.6 seconds, and the timeline readout agreed at 00:03:17. Its graph sits at 10x for most of the clip and dips to roughly 1x for a short stretch in the middle. In the exported file the picture changes about twice as much from frame to frame at the two ends as it does through the middle, which is the shape the graph describes.
The same circle after the Bullet curve: it whips round, holds near the halfway point, and whips out. Shown here as frames in sequence; the full video and a GIF version are supplied alongside this document.
Whether that reads better than the even orbit depends on where the clip is going. The compressed version is quick and loses the slow reveal of the street. The even version is easier to cut into, and the curve can be softened by dragging the points on the graph. Either way, judge it on one question before anything else: watch the subject alone, ignoring the background, and check that no part of him moves between the first frame and the last.
Written 4 September 2026. The man, the street and all three clips were generated in CapCut during a single session and depict no real person or place; the clothing carries no brand marks. Interface labels, timings and generation settings reflect that session and may change.