AI Video Prompts for Slow Motion
Cinematic impact moments, sports action frozen in time, liquid pouring at 240fps, and fabric catching air in a single extended beat — these Seedance 2.0 slow motion prompts direct the model to generate footage where time itself becomes the visual subject.
Slow motion is not a speed setting — it is a directorial decision about which moment in the physical world deserves to be expanded. The wrong moment in slow motion produces footage that looks like a timing mistake; the right moment produces the single frame that defines the entire clip. The structural problem with most AI slow-motion prompts is that they apply the slow-motion instruction to the entire scene rather than to the specific physical event that earns it. "Slow motion of a car crash" distributes the temporal expansion evenly across approach, impact, and aftermath — and the result is footage where nothing peaks because nothing contrasts. "A car impact: real-time approach at full speed, then time dilates on the instant of collision — the front panel crumpling frame by frame, glass fracturing outward, the body absorbing the force across two seconds of expanded time, then returning to normal speed as the car comes to rest" is a complete slow-motion direction because it specifies where normal time ends, what the expansion reveals, and where time returns. The event that earns slow motion shares a common structure across every genre: a moment of peak energy transfer. In sports, it is the contact between ball and bat, foot and ball, fist and jaw — the instant where kinetic energy crosses from one object to another. In nature, it is the water impact, the wing-beat at maximum extension, the predator's strike at the moment of reach. In fashion, it is the fabric's maximum air-catch at the peak of a turn — the point where textile physics are most visible. In drama, it is the tear that finally breaks the surface of the skin, the hand that releases the object, the expression that passes across the face in the half-second before composure returns. Identifying the peak energy transfer in your scene is the first decision of slow-motion direction, and it belongs in the prompt as a spatial and temporal specification: "time slows to maximum extension at the moment of contact, frame-by-frame detail on the impact surface, then returns to real time as the energy dissipates." Frame rate specification is the second layer. The human eye processes approximately 24 frames per second; cinema runs at 24fps; sports broadcast at 50-60fps. Slow motion multiplies this frame rate and then plays it back at 24fps, stretching time by the multiplication factor. Naming the frame rate in the prompt gives Seedance a compression ratio to execute: 120fps played at 24fps is 5× slow; 240fps is 10× slow; 1000fps (available on specialist cameras for ultra-slow motion) stretches a single second of action into 40 seconds of screen time. These are different effects — 5× slow is dramatic; 10× slow reveals mechanics the eye cannot see in real time; 1000fps reveals fluid physics that are physically invisible at human time scales (the Worthington jet from a raindrop impact, the pressure wave in a flexing bat, the way a soap bubble collapses inward before the surface ruptures). Naming the frame rate gives the model the scale of temporal expansion and therefore the level of physical detail to render. Physical material behavior in slow motion differs from real-time behavior in ways that AI video needs to be directed to render correctly. Water in slow motion follows different visual physics than water at real speed: at 120fps, a splash breaks into discrete droplets with visible surface tension at each one; at 240fps, the Worthington jet that rises from the center of an impact has time to form, extend, and collapse; at 1000fps, the water surface dome that briefly forms before rupture is visible as a distinct phase. Without naming these physics explicitly, AI video defaults to generic slow-motion water that looks like sped-up footage run backward — the droplets are too large, the motion too linear. Naming the specific phenomenon — "the Worthington jet at the center of the droplet impact, rising and collapsing over 2 seconds of expanded time, each droplet visible as a discrete sphere with surface-tension roundness" — gives Seedance the specific physics to render. Cloth and hair in slow motion are the two material systems most commonly misrendered by AI video. Real fabric in slow motion shows secondary motion — the main body of the fabric leads the edge by one-half to a full beat, and the lighter the material, the longer the lag. Silk at 240fps in a full turn shows the body of the dress at the peak of the rotation, then the hem completing the arc one to two frames later; the chiffon panel at the shoulder catching air and extending outward like a secondary wing. Without naming these material-specific physics, Seedance defaults to fabric that moves as a rigid surface — which is the visual tell of AI-generated fashion and dance video. Naming the material lag explicitly — "the hem lagging the turn by half a beat, extending outward as inertia continues the arc after the body has stopped" — gives the model the physics grammar of real fabric in slow motion. The camera's role in slow motion is to earn the right to slow down — meaning that the approach to the peak moment should be in real time, at normal speed, so that the slow-motion expansion is felt as a tempo change rather than as the baseline state of the scene. A clip that opens in slow motion and stays in slow motion loses its temporal contrast; a clip that opens at real speed, slows at the peak moment, and then either holds slow or returns to real speed has earned its temporal expansion through contrast. This real-time-to-slow-motion-to-real-time structure is the narrative grammar of slow motion, and it belongs in the prompt: "approaches at full speed, the moment of impact dilates to slow motion, holds at the lowest frame rate for the peak contact, then time expands back to real speed as the aftermath resolves." That temporal arc is the whole architecture of effective slow-motion direction, and naming it explicitly is what gives Seedance a complete brief rather than a speed instruction.
More use cases
Frequently asked questions
What are the best AI video prompts for slow motion?
The best slow motion prompts name three things: the specific peak moment that earns the expansion ("the instant of ball contact," "the fabric at maximum air-catch," "the drop breaking the water surface"), the frame rate that defines the scale of the expansion ("240fps," "1000fps ultra-slow," "120fps played at 24fps"), and the temporal structure — when real-time becomes slow motion and whether it returns. Without all three, the model defaults to footage that moves slowly throughout rather than footage where a specific moment earns its temporal weight. Every prompt in this gallery uses that structure.
How do I write a slow motion prompt for Seedance 2.0?
Structure the slow motion prompt in three layers: (1) the real-time approach — name what is happening before the peak event, at full speed, so the model has context for the tempo change; (2) the expansion trigger — the specific physical moment where time slows, named as precisely as possible ("the instant the bat contacts the ball," "the moment the water surface breaks," "the fabric at its fullest extension"); (3) the frame rate and material behavior — "240fps, water breaking into discrete surface-tension droplets, the Worthington jet rising from the center" or "120fps, silk hem lagging the turn by one beat, extended outward by inertia." The real-time approach is the most commonly omitted element, and it is what makes slow motion feel earned rather than applied.
Can Seedance 2.0 generate realistic slow motion video?
Yes. Seedance 2.0 handles slow motion well when the prompt names the specific physical behavior of the materials at the expansion moment. Slow motion water, fabric, and hair each have distinct physics that differ from their real-time behavior — naming them explicitly ("discrete water droplets with surface-tension roundness," "fabric hem completing the arc after the body stops," "hair secondary motion lagging the main movement by one beat") produces slow motion that reads as physically real rather than as video that has simply been slowed down. Browse the slow motion prompts here for examples with preview videos.