Motion is the hardest thing to get right in a video prompt: describe it well and your footage feels alive; describe it vaguely and subjects smear, limbs multiply, and the camera drifts with no intent. Most "bad AI video" isn't a resolution problem, it's a motion problem. This guide breaks down how to prompt action and movement convincingly, and how CoreReflex scores motion coherence on every shot so the failures regenerate instead of shipping.
Why motion is the hardest part of a video prompt
A still image only has to be plausible for one frame. Video has to stay plausible across dozens of them while everything moves in a way that obeys physics and intention. That's a much taller order, and it's where text-to-video models struggle most: a hand passing behind a back, a car turning a corner, hair moving in wind. When a model loses the thread, you get the classic artifacts, warping, ghost limbs, objects that pop in and out, a camera that floats without purpose.
The fix starts in language. Models render what you specify, and "a man walking" leaves almost everything undetermined. The more precisely you describe the motion, what moves, how fast, and how the camera relates to it, the less the model has to guess, and the fewer guesses it gets wrong.
The anatomy of a motion description
A strong motion prompt separates three layers and is explicit about each. (For the broader structure of a prompt, our anatomy of a great AI video prompt breaks down all the components; here we focus on the moving parts.)
Subject action
Name the action with a strong, specific verb and a direction. "A barista slides a cup across the counter toward the camera" gives the model a trajectory; "a barista with a cup" gives it nothing. Specify what initiates and what completes the motion, start pose, end pose, so the model has a path to interpolate, not a vague vibe.
Camera movement
Decide whether the camera is still or moving, and say so explicitly: a slow push-in, a handheld follow, a locked-off tripod shot, a crane up. Tie the camera to the subject, "the camera tracks left to stay with her as she walks", so subject and camera motion reinforce each other instead of fighting. CoreReflex passes these moves through to the engines that support them: Kling honors structured camera control, and Veo takes camera direction in the prompt.
Speed and continuity
Pacing is part of the motion. "Slow, deliberate" reads completely differently from "quick, snappy." State the tempo, and for a multi-shot sequence, think about how one shot hands off to the next. CoreReflex anchors the last frame of a shot as the starting point for the next, so a continuous action, someone reaching for a door, the door opening in the next shot, actually stays continuous across the cut.
How to prompt motion and action, step by step
- Pick one primary action per shot. Don't ask a single shot to do three things. One clear movement renders far more reliably than a crowded one.
- Use directional verbs. Replace static descriptions with verbs that imply a vector: glides, lunges, spins, drifts, accelerates.
- Set the camera relationship. State whether the camera holds, follows, pushes, or orbits, and at what speed.
- Specify start and end. Give the motion a beginning and an end state so the model interpolates a coherent path.
- Name the tempo. Slow and weighty, or fast and kinetic, pacing shapes how the motion reads.
- Plan the handoff. For sequences, describe how the action continues into the next shot so continuity holds.
These steps slot directly into the broader workflow in our text-to-video prompting guide, and they pair well with the look-and-feel choices covered in film-style prompts for a cinematic look.
Common motion failures and how to phrase around them
| Symptom | Likely cause | Prompt fix |
|---|---|---|
| Warping / morphing subject | Too much motion, underspecified path | One action, explicit start and end pose |
| Ghost or extra limbs | Fast, ambiguous movement | Slow the tempo; describe limb position |
| Floating, aimless camera | No camera intent stated | Name the move and tie it to the subject |
| Jump between shots | No continuity anchor | Describe how the action carries over |
Phrasing around the failure is faster than fighting it after the fact, but you won't always catch it by eye, especially at volume. That's what the quality gate is for.
What motion coherence means, and what happens when a shot fails
Motion coherence is whether the movement in a shot is physically and temporally consistent, subjects move along believable paths, the camera moves with intent, and nothing warps, duplicates, or stutters between frames. It's one of the concrete checks CoreReflex scores on every generated shot, alongside prompt match, sharpness, on-screen text legibility, on-brand, and claims-risk.
When a shot fails the motion-coherence check, CoreReflex doesn't make you start the whole film over, it selectively regenerates the failing shot, re-rolling just that clip until the movement holds, while the shots that already passed stay untouched. That's the difference between a guess-and-check toy and a production tool: bad motion gets caught and fixed automatically instead of slipping into your final cut.
Let the quality gate do the QA
Eyeballing every frame of every shot doesn't scale, and motion artifacts are easy to miss on a fast first watch. Scoring motion coherence per shot turns "hope it looks right" into a measured gate. And because every generation carries a portable provenance trace, model, prompt, parameters, and score, a shot that nails a tricky action is reproducible: you can replay the exact recipe instead of trying to re-describe the magic. To dive deeper across formats and tactics, browse our AI Video guides, and the documentation covers how scoring and regeneration work under the hood.
Frequently asked questions
How do I prompt motion in AI video?
Name one primary action per shot using a directional verb, set the camera's relationship to that action, and specify the start and end of the movement plus its tempo. The more precisely you describe what moves and how, the less the model has to guess, and guesses are where artifacts come from.
What is motion coherence?
Motion coherence is whether movement in a clip stays physically and temporally consistent: subjects follow believable paths, the camera moves with intent, and nothing warps or duplicates between frames. CoreReflex scores it as one of the concrete checks in its quality gate on every shot.
What happens if a shot's motion looks wrong?
The shot fails the motion-coherence check and is selectively regenerated, only that clip re-rolls, while shots that already passed are left alone. You get a clean replacement without rebuilding the whole film.
Can I control the camera movement, not just the subject?
Yes. State the camera move explicitly, push-in, follow, orbit, locked-off, and CoreReflex passes it to engines that support it, including Kling's structured camera control and Veo's prompt-based direction.
Put motion into words the model can render
Convincing AI video is a writing problem before it's a rendering problem: describe the action, the camera, and the tempo precisely, and let the motion-coherence check catch what slips through. Start free, no credit card, and prompt your first moving shot in a sentence.