The camera is dollying forward, the subject is running, the background is swirling, the leaves are blowing, the hair is flowing, the curtain is flapping. Every element in the frame is in motion. The viewer’s eye doesn’t know where to land. The clip feels chaotic instead of cinematic.
Real cinematography uses motion sparingly — usually one element moves at a time so the eye can follow. AI video defaults to “make everything move” because static-looking clips score lower in user feedback. You have to fight that default.
Common causes
Ordered by hit rate, highest first.
1. Motion strength too high
Most tools default motion to a middle-high preset. At high motion, the model adds movement to everything in frame.
How to spot it: motion slider is at default or above mid. Drop it.
2. Multiple action verbs in the prompt
"a woman runs through a field, waving her arms, hair blowing, leaves swirling, wind rushing"
Each verb adds motion. Cumulative effect = chaos.
How to spot it: count motion verbs in prompt. >2 = likely too much.
3. Camera + subject + background all moving
"slow dolly forward through a busy market crowd, neon signs flickering"
Camera moves. Subjects move. Lights flicker. Three layers of motion compete.
How to spot it: prompt describes motion at multiple layers (camera, subject, environment).
4. Background motion adjectives
swirling, rushing, flickering, flowing, dynamic, chaotic, bustling — these are pure motion adjectives. Each adds energy.
How to spot it: prompt has these descriptors.
5. Source image has implied motion (image-to-video)
If you upload a motion-blurred or dutch-angle source, the model assumes you want continued motion energy.
How to spot it: source image looks like a motion still rather than a calm photo.
6. Style implies motion-heavy aesthetic
action movie, music video, parkour, fast cuts, dynamic — these style anchors push toward kinetic motion.
How to spot it: style word evokes high-energy aesthetic.
Shortest path to fix
Step 1: Drop motion strength to minimum
# Runway: 1-2 (was 5+)
# Pika: 0.2-0.3 (was 0.5+)
# Kling: "subtle" (was "intense")
# Luma: low (was high)
This single change fixes 50%+ of over-motion cases.
Step 2: Strip the prompt to ONE action verb
# Bad — overloaded
"a woman runs through a field, waving her arms, hair blowing, leaves swirling, wind rushing"
# Good — one verb
"a woman slowly turns her head to look at the camera"
# Or another single verb
"a woman walks forward at a steady pace"
Pick the most important action and delete everything else.
Step 3: Pick ONE moving layer
Decide: camera, subject, or background. Only one.
# Moving camera, static subject
"slow dolly forward, subject stays perfectly still"
# Static camera, moving subject
"static locked camera on tripod, no pan or zoom, subject walks forward"
# Subtle background, both else static
"locked camera, motionless subject, only the leaves in background gently sway"
Step 4: Delete background-motion adjectives
Remove from prompt:
swirling, rushing, flickering, flowing, dynamic, chaotic,
bustling, energetic, vibrant motion, kinetic, action-packed,
high-energy, lively
Replace with calm descriptors if needed:
still, calm, peaceful, locked, quiet, gentle, settled
Step 5: Use a static-looking start frame
For image-to-video:
# Good start frame characteristics
- Subject in a settled, stable pose
- Sharp focus, no motion blur
- Level horizon, no dutch angle
- Centered composition
# Avoid
- Motion-blurred source
- Subject mid-action
- Dynamic angle
- Off-center composition
Step 6: Strip motion-heavy style words
Remove:
action movie, music video, parkour, fast cuts, dynamic,
high-octane, kinetic
Replace with calm style:
slow cinematic, documentary, observational, steady, contemplative
Prevention
- Storyboard before generating: decide ONE thing that moves
- Default motion strength to minimum; raise only when needed
- Default to one verb per prompt; multiple verbs = multiple motions
- Pick “calm” style anchors unless the clip specifically needs energy
FAQ
Q: Why does the model add motion to things I didn’t ask to move? A: Static-looking clips score lower in human-feedback training, so models learn to add ambient motion (swaying leaves, fabric ripple, micro camera drift) by default. You have to explicitly suppress motion on every layer you want still — “locked camera, motionless subject” beats hoping the model leaves things alone.
Q: Is low motion strength the same as a still image? A: No. Minimum motion still produces a video — subtle breath, eye blink, slight subject sway. It just removes the chaotic-everything-moves default. If you want a true still, use a still-image tool or a 1-frame video.
Q: Why does dropping motion strength sometimes look worse — robotic or frozen? A: Some tools quantize the motion slider in coarse steps, so “minimum” can clip past natural micro-motion into uncanny stillness. If that happens, raise one notch and trim verb count instead — the goal is one coherent motion, not zero motion.
Q: How many motion verbs is “too many” in a prompt? A: For most tools, more than two motion verbs (running + waving + blowing) compounds into chaos at typical motion-strength defaults. One primary verb + one ambient detail (“walks forward, leaves gently sway”) is usually the ceiling for coherence in a 4–10s clip.