Prompt craft · video
How to Write AI Video Prompts in 2026 (Motion, Camera & Templates)
Image prompts pasted into video jobs waste credits. Motion models need action, camera, duration, and rejects as separate blocks — and they punish wardrobe rewrites when the plate was already right. This guide is the video sibling of our stills prompt playbook.
Quick answer: six-line video brief
If a line can fail on its own, keep it on its own.
- 01Subject: one or two clearly adult identities with stable facts.
- 02Action: one verb phrase (lean in, slow embrace, turn toward camera).
- 03Camera: shot size + move (static / slow push / gentle arc).
- 04Light + finish: match the plate or name the change once.
- 05Duration / motion tier: short + low on first paid run.
- 06Rejects: extra limbs, identity swap, wardrobe crawl, underage cues.
Image briefs vs video briefs
Stills care about wardrobe edges and hands in one frame. Video adds time: contact frames, temporal flicker, and camera paths that invent background.
| Block | Stills emphasis | Video emphasis |
|---|---|---|
| Subject | Identity + wardrobe lock | Same — change only if the plate must change |
| Action | Pose noun | Verb + intensity over time |
| Camera | Shot size / lens feel | Shot size + move + hold |
| Rejects | Fingers, logos, age cues | Those plus flicker, morph, wardrobe crawl |
Core formula for review-ready video prompts
Keep the order boring: subject → action → camera → light → duration/constraints → rejects. Excitement belongs in the authorized performance you already have rights to, not in adjective piles.
- 01One primary action. Secondary motion only after the first run holds.
- 02Name adults as adults. No age-ambiguous slang.
- 03If you attach a still, treat wardrobe as locked unless the job is a deliberate restyle.
Motion tiers and examples
Preflight checklist before any motion spend.
| Tier | Example brief fragment | Use when |
|---|---|---|
| Low | soft breath, hair drift, subtle smile, static camera | First i2v on a hero still |
| Medium | slow lean-in embrace, gentle head turn, slow push-in | Dual-photo hug/kiss after framing matches |
| High (defer) | fast spin, complex hand choreography, wardrobe change mid-clip | Only after low/medium identity holds — often wrong for private plates |
Text-to-video templates (skeletons)
Fill in the brackets with the subject, action, camera, and lighting. Keep each short clip focused on one visible action.
| Job | Skeleton |
|---|---|
| Portrait motion | Adult [identity], [wardrobe], looking to camera, [low action], [shot size], [camera move], [light], avoid [rejects] |
| Approach / lean-in | Two adults [A] and [B], matched framing, slow lean-in, soft practical light, medium close-up, gentle push-in, avoid hand melt and identity swap |
| Embrace hold | Two adults in a slow embrace, minimal sway, faces readable, static camera, short duration, avoid wardrobe crawl |
Image-to-video: prompt only what should move
When the still is approved, the prompt’s job is motion and camera, not a second cast sheet. Restating a new haircut or outfit invites drift.
- 01Reference the plate implicitly: “keep identity and wardrobe; add …”
- 02Prefer low tier verbs on run one.
- 03If contact needs a second person, switch to dual-photo kiss/hug instead of inventing a body.
Camera and lens mini-table
| Intent | Phrase to try | Watch-out |
|---|---|---|
| Hold beauty | static medium close-up | Boring is good on run one |
| Intimate push | slow push-in | Too fast → face morph |
| Reveal space | gentle arc left/right | Background invent / flicker |
| Documentary | handheld micro-shake | Can amplify hand errors |
Settings that actually matter
- 01Duration — shorter first; extend only after review.
- 02Motion strength / creativity — start low on private plates.
- 03Seed / reuse controls — keep when comparing one prompt line.
- 04Cost estimate — read it; trust Studio over blog prices.
Common mistakes that burn video credits
- 01Pasting a still megaprompt unchanged into a video box.
- 02Asking for wardrobe change + new camera + new action in one run.
- 03Using celebrity names as free identity.
- 04Skipping the full-resolution pass before a second paid clip.
- 01Adults read clearly adult in every sampled frame.
- 02Action matches the single verb you wrote.
- 03Camera move matches the brief.
- 04Identity holds vs the plate or dual source photos.
- 05Rejects list only true failures.
Privacy, consent, and honest limits
Private history is not the same as local-only. Upload only media you have rights to use. Hard policy blocks stay hard — prompt craft is not a bypass.
See Acceptable Use. For contact workflows, prefer kiss / hug tools after the brief is ready.
Preflight checklist before any motion spend
Print this mentally. If any line fails, do not generate video yet.
- 01Every adult in frame authorized this use; no minors; no non-consensual intimate fakes of real people.
- 02Still is sharp at face, hands, and wardrobe edges at 100% zoom.
- 03Composition leaves room for the motion you will ask for (no cropped elbows if hands will move).
- 04Identity anchors written down: hair, marks, wardrobe — same words you will reuse in the motion brief.
- 05On-screen credit / duration estimate read before click.
- 06Export plan is private history first — not a public gallery toy.
FAQ
Separate subject, action, camera, light, duration, and rejects. Change one failing line per paid rerun.
Yes. Keep visual facts, but add motion and temporal rejects. Do not paste still paragraphs unchanged.
Prefer visible action and camera facts. Graphic prose helps chat/fiction models more than motion models and can increase refusals.
Low on private plates. Increase only after identity holds.
i2v for small motion on one plate; dual-photo when near-contact between two people is the point.
Next step
Test one motion brief privately
Lock identity, write a six-line video brief, pick low motion, generate once, and only then edit the failing line.
