AI AI Tools Guidebook
Prompt Library

Character Motion Video Prompts: 10 Templates That Don't Break Hands

10 character-motion AI video prompts with explicit start + end poses to keep limbs intact — plus which 2026 model (Kling 3.0, Veo 3.1, Runway Gen-4.5) handles each.

Published: Updated: Author: AI Productivity Guide Team 🌐 查看中文版本

Character-motion clips rarely fail on the motion itself. They fail on pose stability at the start and end of the shot, where a hand folds in on itself or a foot melts into the floor. The fix is structural, not magic: write every prompt as start pose + single action + end pose, and let the camera stay still. Below are 10 templates built that way, plus which June 2026 model to run each on.

TL;DR

  • Always specify three things: where the body starts, the one thing it does, and where it ends.
  • Keep the camera static and pick one light. Subject motion plus camera motion compounds errors.
  • 4–6 second clips are the stability sweet spot. Most second-half deformation starts after 6s.
  • Avoid fine hand work (typing, piano). For human motion, Kling 3.0 is the strongest value pick; Runway Gen-4.5 wins on control (keyframes + motion brush); Veo 3.1 wins on prompt adherence and lets you lock a face with up to 4 reference images.

What a high-quality prompt should contain

Five layers every character-motion prompt needs:

  • Start pose: starts standing left of frame / crouched low
  • Action: a single move with explicit pace (walks at normal pace / slow gentle rise)
  • End pose: exits frame right / holds position
  • Lens + lighting: static medium shot, soft golden hour light — both fixed
  • Duration: 4–6s is the sweet spot

The English prompt text works the same across Kling, Veo, and Runway — you write the same descriptive sentence and paste it into whichever tool you’re using.

Which model for which motion (as of June 2026)

ModelBest atBase clipEntry priceNote
Kling 3.0Complex human motion (walk, dance, fabric, hair)up to 10s$6.99/mo StandardBest value for body motion; strong from a reference still
Runway Gen-4.5Control: first/last keyframes, motion brush, character consistency~10s$15/mo StandardTops the Artificial Analysis text-to-video benchmark
Veo 3.1Prompt adherence, native audio, 4K, identity lock8s (extendable)API ~$0.15–0.40/sec; in Google AI Pro $19.99/moUpload up to 4 reference images to keep a face/outfit consistent

A note on Sora: OpenAI discontinued the Sora web and app on April 26, 2026, with its API ending September 24, 2026, so these prompts target the three models above.

10 copy-ready prompt templates

1. Walks past camera

Best for: B-roll, character entrance · Run on: Kling 3.0

A young woman starts standing left of frame, walks across at normal pace from left to right, exits frame right, static medium shot, soft golden hour light, 5 seconds

2. Turns and smiles

Best for: Brand film, emotional beats · Run on: Veo 3.1 (lock the face with a reference image)

A man stands facing away from camera, slowly turns his head over his right shoulder toward the camera, gentle smile, then holds position, static shot, 4 seconds

3. Dancer rise motion

Best for: Sports / dance brands · Run on: Kling 3.0

A young dancer in athletic wear begins in a low crouch, smoothly rises and extends arms outward in a single breath-paced motion, static medium shot, soft window light, 5 seconds

4. Barista pour

Best for: Coffee brands, lifestyle video · Run on: Veo 3.1 (liquids + native steam)

A barista at a counter pours steamed milk into a coffee cup, single smooth pour, gentle steam, warm cafe lighting, slight slow-motion, static medium shot, 6 seconds

5. Child looks up laughing

Best for: Family / education brands · Run on: Kling 3.0

A child laughing and looking up at falling autumn leaves, slow gentle giggle, single static medium shot, warm afternoon light, shallow depth of field, 5 seconds

6. Craftsman tap

Best for: Handmade / premium brands · Run on: Runway Gen-4.5 (motion brush on the hammer)

An older craftsman in a workshop slowly raises a hammer and taps a metal piece twice, focused expression, warm tungsten light, static slight low angle, 5 seconds

7. Runway walk

Best for: Fashion week, brand visuals · Run on: Runway Gen-4.5 (keyframe the start and end pose)

A model in evening wear walks slowly toward the camera on a polished runway, single smooth gait, soft top-down studio light, anamorphic 35mm, 6 seconds

8. Skater push-by

Best for: Athletic / streetwear brands · Run on: Kling 3.0

A teenager skateboarder pushes off and rolls past camera once, ground-level tracking shot, golden hour light, 35mm lens, slight motion blur on wheels, 5 seconds

9. Anime sword draw

Best for: Anime intro shot · Run on: Kling 3.0 or Runway Gen-4.5

A young swordsman draws his katana from sheath in a single fluid motion, dust kicks up at feet, slight slow-motion, hand-drawn 2D cel-shaded anime style, 4 seconds

10. Scientist examines beaker

Best for: Pharma / tech brands · Run on: Veo 3.1 (clean prompt adherence, glass + reflections)

A scientist in a lab coat slowly holds up a glass beaker, examines it under bright overhead light, single subtle hand rotation, clean lab background, 5 seconds

Common mistakes

  • Two distinct actions in one clip — walks over and turns almost always fails. One action per clip.
  • Action without a start / end position — the model improvises the in-between, usually badly.
  • Moving the camera too — subject motion plus camera motion compounds errors. Lock the camera first.
  • Asking for fine hand work (piano, typing) — fingers are still the weakest area in mid-2026 models; expect breakage.
  • Going past 6s — the second half deforms even on Kling 3.0 and Veo 3.1. Generate short, extend or cut in post.

How to push results further

  • Static camera + moving subject = the most stable combo across all three models.
  • For slow-mo, slight slow-motion, breath-paced is more stable than a bare slow motion.
  • Expression beats large motion: gentle smile, soft eyes survives better than a sweeping pose change.
  • Name where a prop sits in the hand (cup held in right hand) — unanchored props drift.
  • For a recurring character, use Veo 3.1 reference images or Runway Gen-4.5 character consistency so the face doesn’t change between clips.
  • Need a longer sequence? Generate 4–6s beats and cut them together, or use Veo 3.1’s scene extension rather than asking for one long take.

FAQ

Q: Subject deforms while walking past camera — how do I fix it?

A: Swap normal pace for slow steady pace and add single smooth gait. Slower motion gives the model fewer frames to invent, so limbs stay intact. Kling 3.0 handles this walk most reliably.

Q: Fingers always look broken — what works?

A: Avoid hand close-ups entirely. If a hand must be visible, add hands visible with five distinct fingers and keep the hand still rather than performing fine work. Fine finger articulation is still the weakest area across Kling 3.0, Veo 3.1, and Runway Gen-4.5 as of June 2026.

Q: How do I keep the same character across several clips?

A: Use Veo 3.1’s reference images (up to four — face, outfit, style) or Runway Gen-4.5’s character consistency feature. A plain text prompt alone will drift the face between generations.

Q: Can I do two people in conversation in one shot?

A: Multi-person scenes are still unreliable. Shoot two single-person clips and cut them as split-screen or shot/reverse-shot. This is far more stable than asking one prompt for both people.

Q: Expression transitions look stiff — why?

A: Expressions count as motion, so do only one transition per clip. Split it: neutral expression first, smile in a second clip, then cut.

Q: Which model should I start with?

A: For most character motion on a budget, start with Kling 3.0 ($6.99/mo Standard). Choose Runway Gen-4.5 when you need keyframe or motion-brush control, and Veo 3.1 when prompt adherence, 4K, or a locked face matters most.

Tags: #Video generation #Prompt #Cinematic

Related Articles