How to write prompts for subtle motion

Published: 2026-03-13 Updated: 2026-03-31 By: Seho Jung

Answer-first summary

Short prompts with one subject and one motion create the most stable results. Use gentle language (“slowly”, “soft light”) and avoid stacking multiple actions.

Many users try to fix bad results by adding more words. In practice, clarity beats length. A short prompt that clearly describes a single motion will often outperform a detailed paragraph.

1. Start with a one-line goal

Decide the single motion that matters most. If you can’t summarize the motion in one line, the prompt is probably too complex for a short clip.

2. Ask for one action only

“Smile + wave + camera orbit” creates competing instructions. Start with one action, validate it, and then add a second action only if the first remains stable.

3. Minimize background changes

Background transformations are a common source of artifacts. Keep the background static unless the clip’s purpose depends on it.

4. Use gentle modifiers

Words like “slowly”, “soft”, and “gentle” reduce instability. Strong verbs and aggressive pacing increase the likelihood of warping.

Prompt examples

“Warm light, subject slowly smiles and tilts head slightly.”

“Bright studio, product slowly rotates, subtle camera push.”

5. Keep prompt length in check

If your prompt needs several commas, it might be too complex. Use a single sentence whenever possible and remove any detail that doesn’t directly support the motion.

6. Edit by removing, not adding

When results look wrong, remove parts of the prompt instead of adding new instructions. In most cases, simplification fixes the issue faster.

7. Stable vs unstable examples

Stable: “Soft studio light, product slowly rotates.”

Unstable: “Product spins fast, camera circles, background changes, reflections intensify.”

Expectation setting: prompts are not magic

Prompts cannot fix a messy input image. If the photo is unclear or cluttered, simplify the image first, then refine the prompt.

In short clips, precise language beats dramatic language. Clear verbs and gentle pacing work best.

Reusable prompt templates

A simple prompt revision loop

When results look unstable, remove parts of the prompt before adding more. This loop keeps changes controlled and makes improvements easier to see.

  1. Reduce the prompt to one action.
  2. Remove background transformation requests.
  3. Add a gentle pacing word like “slowly”.
  4. Change one word at a time and test again.

Before/after prompt log example

A prompt like “product spins fast, camera orbits, background changes” often produces unstable motion. Reducing it to “product slowly rotates, soft studio light” usually stabilizes the output.

Keeping a short log of these changes helps you learn what your model responds to most reliably.

Words that improve stability

These gentle modifiers help reduce chaos. If the clip still wobbles after shortening the prompt, try adding one of these.

Conclusion

Prompt quality is about clarity, not complexity. One subject, one motion, gentle pacing. That’s the formula for stable short clips.

Prompt templates

The longer the prompt, the more the goal gets diluted. Start with a minimal template and expand only if needed.

How to adjust motion intensity

Higher motion can look exciting but increases the risk of distortion. Start low and raise intensity only when the output stays stable.

Prompt revision loop

If the result looks off, reduce the prompt before adding more detail. Removing weak instructions often clarifies the model’s focus.

The guide and the article on why results look unnatural explain the most common fixes.

Simplify instead of stacking negative prompts

Listing many “do not” instructions is often less effective than simplifying the goal. Clarity helps the model focus on a single motion outcome.

For example, replacing “no exaggeration, no wobble, no distortion” with “slow gentle smile” often produces a more stable result.

FAQ

Q: Can I add background changes later?
A: Yes, but validate the base motion first. Add changes incrementally.

Q: Are long prompts always bad?
A: Not always, but they increase conflict. Short prompts are more reliable for short clips.

Q: What’s the fastest way to improve a prompt?
A: Remove extra actions and keep only the core movement.

Q: Does it need to be written in English?
A: Clarity matters more than language. Keep it short and specific.

Q: Should I use negative prompts?
A: They can help, but start by simplifying the main prompt first.

Q: Can the same prompt produce different results?
A: Yes. Model behavior can vary, so change one variable at a time.

Related reading