You asked for a video clip and the tool returned what is essentially a still image with a 5-pixel parallax: a tiny camera nudge, maybe one leaf blowing, the rest stone-dead. You wanted action and got a slideshow. This is the opposite of motion jitter or drift — here the model under-committed to motion.
Fastest fix: rewrite the prompt so it contains (1) a concrete action verb on the subject and (2) exactly ONE named camera move, then push the tool’s motion control up one notch. Those three changes recover visible motion in the large majority of cases without triggering drift. Everything below is the longer diagnosis if that does not do it.
Which bucket are you in?
| Symptom | Most likely cause | Go to |
|---|---|---|
| Prompt is a scene description with no verbs | No action for the model to animate | Step 1 |
| Subject moves a little, camera is locked | No camera move specified | Step 2 |
| Some motion but far too timid | Motion control set too low / default is conservative | Step 3 |
| Image-to-video output barely moves | Static reference image + passive prompt | Step 4 |
| Output averages to almost-still | Contradictory motion words in prompt | Step 5 |
| One render moves, the rest are static | Seed/variance, or 2-3s clip is too short | Steps 6-7 |
Common causes
Ordered by what most often produces under-motion.
1. No verbs in the prompt
A woman in a red dress at a cafe is a still-image prompt. There is no action for the model to animate, so most video models fall back to minimal motion when the prompt does not say what should move.
How to spot it: read your prompt and underline every verb. If there is no movement verb (running, turning, waving, and so on), the prompt is implicitly asking for a still.
2. Motion control set too low
Runway’s Motion Brush at 1, Pika’s motion slider near the bottom, Kling’s Motion intensity at a low percentage — these settings are good for avoiding drift but produce near-still output. Real motion needs the setting higher.
3. Tool default favors subtle motion
Most video tools ship conservative motion defaults because aggressive motion produces drift and morphing. As of June 2026, Runway Gen-4 and Gen-4.5 are image-to-video-first and treat the still as the source of truth, so a thin prompt yields a near-static result. Kling and Pika both default toward the safe middle of their motion ranges.
4. Static reference image with passive prompt
In image-to-video, if the reference image is a perfectly still subject (person standing straight, product centered) AND the prompt does not name a motion, the model reads “static” as the cue and holds the frame.
5. Camera motion not specified
The model defaults to a locked camera unless you specify otherwise. If neither subject motion nor camera motion is specified, the output is a near-still.
6. Conflicting motion descriptors
Slow gentle movement, dynamic action, intense motion — three contradictory motion descriptors. The model averages them toward “very little motion.” The same applies to camera moves: stacking two or three in one prompt confuses the model and it often defaults to almost no camera travel rather than picking one.
7. Clip duration too short to develop motion
A 2-second clip rarely shows much movement; the model knows it does not have time for an action arc. 5-second clips (the standard short tier on Runway Gen-4 and Kling) are far easier to fill with visible motion.
Before you change anything
- Save the prompt, model/version, motion settings, and the static output.
- Decide what kind and how much motion the use case actually needs.
- If using image-to-video, look at the reference image and ask whether it suggests motion at all.
- Back up the prompt template before changing it.
Information to collect
- Full prompt, motion setting value, model, version.
- Clip duration.
- Whether the reference image (if any) shows a subject in motion or at rest.
- The intended motion type and intensity.
Shortest path to fix
Step 1: Add concrete action verbs
Replace the descriptive scene with an actionable scene.
Before:
a woman in a red dress at a cafeAfter:
a woman in a red dress, picking up a coffee cup, taking a sip,
glancing toward the windowStrong verbs the model can act on: running, walking, turning, looking, waving, smiling, laughing, pouring, opening, closing, reaching, leaning, standing up, sitting down. For fast action, escalate the verb itself: sprinting, leaping, soaring, shattering carry more motion than moving or going.
Step 2: Specify ONE camera move
Even a subtle camera move transforms a static-feeling clip — but name only one per shot. Stacking moves is a top cause of the model giving up and locking the camera (see cause 6).
slow camera push-in toward the subjectslight pan from left to right across the scenearc shot around the product, 30 degrees of rotationUse real cinematography terms the models are trained on: dolly in/out, push-in, pull-back reveal, pan, tilt, crane up, orbit, tracking shot. On Runway Gen-4/4.5 in image-to-video mode the text prompt should be almost entirely about motion (the image already carries subject, composition, color and style), so spend those words on the camera move and the action.
Step 3: Increase motion strength
Find the right control for your tool and push it up one notch:
- Runway (Gen-4 / Gen-4.5): raise the Motion Brush strength on the region you want moving; lower the Motion Bucket only if you get distortion. Defaults are conservative.
- Pika (2.2 / 2.5): move the motion slider up; in 2.5, camera direction is a first-class prompt input, so state the move in words too.
- Kling (2.5 / 2.6 / 3.0): the control is now a Motion intensity percentage, not the old
subtle / medium / intenselabels. Use ~60-70% for cinematic motion, 80-90% for energetic / social-media motion. - Google Veo (3 / 3.1): no manual motion slider; it follows the prompt closely, so put the motion entirely in the verbs and the single camera move.
If drift appears at higher motion, treat it as the tradeoff or use a shorter clip at the higher setting.
Step 4: Use a more energetic reference image
For image-to-video, choose a reference that already implies motion:
- Subject leaning forward instead of standing straight
- Hair slightly windblown
- One foot raised
- Hand reaching toward something
The model continues from the reference’s implied motion vector, so a reference that is already “mid-action” produces far more movement than a posed, centered shot.
Step 5: Use motion-specific keywords (and cut the contradictions)
Some prompt phrases reliably increase motion:
dynamic motion, energetic movement, expressive gesture, visible actionDo not combine them with contradictory modifiers (subtle, gentle, slow, still) — the model averages the conflict down to almost no motion.
Step 6: Choose a tool with stronger motion (June 2026)
If your tool fights you on every prompt, switch:
- Kling 3.0 is widely rated strongest for motion realism, especially full-body and action.
- Google Veo 3.1 has the best complex-prompt following (and is the mainstream model with synchronized audio), so motion you describe tends to actually happen.
- Runway Gen-4.5 leads on cinematic camera control once you give it explicit moves.
- Pika 2.5 is the speed pick for energetic social-media clips.
- Note: OpenAI Sora is no longer a viable choice — the web and app were discontinued on April 26, 2026, and the API shuts down on September 24, 2026. Do not build a workflow around it.
Step 7: Extend clip duration
Switch from 2-3 second clips to the 5-second tier (the standard short duration on Runway Gen-4 and Kling). The model allots more motion budget to longer clips. Trim back to 2-3s in post if needed.
How to confirm the fix
- Watch the clip; the subject and / or camera should visibly move.
- Pause at the first and last frames; they should be noticeably different.
- The clip should feel like a “video,” not a “photo with parallax.”
- Re-render three times at the same settings — all three should have motion, not just one lucky output. If only one in three moves, the motion setting is still too low or the prompt is borderline; push one notch further.
If it still fails
- Re-render with both a subject verb AND one explicit camera move named.
- Run the exact same prompt on a different model (Kling 3.0 or Veo 3.1) — some tools simply default very statically.
- Add
cinematic motion, dramatic action sequenceand re-render. - For fast action specifically (running, jumping), use a model trained well on action motion (Kling 3.0, Runway Gen-4.5).
- Package the prompt, model, version, motion settings, and the static output before asking for community help.
FAQ
Why does my image-to-video clip barely move even with a good prompt? In image-to-video the model treats the still as the source of truth. If the reference is a posed, centered, perfectly still subject, “hold the frame” wins over a thin text prompt. Pick a mid-action reference (Step 4) and put the entire text prompt on motion (Step 2).
Is more motion strength always better? No. Past a point you trade static for drift and morphing (warping faces, melting backgrounds). Raise the setting one notch at a time and stop when you have enough motion. Shorter clips tolerate higher motion settings.
Why does only one in three renders have motion? The motion setting is at the borderline where seed variance decides. Push the motion control one notch higher, or add a stronger action verb, so motion stops depending on a lucky seed.
I used to use Sora — what should I switch to? Sora’s web and app shut down on April 26, 2026 and the API ends September 24, 2026. For motion, Kling 3.0 and Google Veo 3.1 are the strongest current replacements; Runway Gen-4.5 for cinematic camera work.
Can I add camera motion to a clip that already rendered static? Yes — several tools offer video-to-video editing that re-renders an existing clip with an added camera move (slow push-in, orbit, pull-back). It is usually faster than re-prompting from scratch, but still name only one move.
Prevention
- Always include action verbs plus one camera move in video prompts; treat both as required.
- Default motion controls to the middle of the range, not the bottom — only drop lower when drift is a real problem.
- Start clips at the 5-second tier; trim shorter in post if needed.
- For image-to-video, pick reference images that already suggest motion direction.
- Standardize on a tool whose motion strength matches your typical use case.