Before/After Transformation Prompts (2026)
The before/after clip is the most honest format on short video: one subject, two states, one cut. And Q4 is transformation season — costume reveals on 10⁄31, gear upgrades on 11⁄27, holiday home glow-ups through 12⁄25, “new me” posts on 1⁄1. Each is a before/after waiting to be shot.
AI video can finally deliver it. Generated transformations used to die on identity drift — the “after” person wasn’t the “before” person. Seedance 2.5, Kling 4.0, Sora 2 and Veo 3 now hold a subject across two states, as long as the prompt defines the relationship between the halves, not just the scenes.
That’s where most prompts fail. Here are the two structural formats, seven shot elements, and three templates you can run today.
Two scenes are not a transformation
A before/after is a comparison with a spine. Three rules hold it together:
- Everything matches except the state. Same subject, same framing, same camera height. Move the camera between halves and viewers read two separate videos.
- The transition is the beat. Hard cut on action, a hand over the lens, a snap, a shutter. Name it in the prompt, or a dissolve kills the illusion.
- The “after” gets stillness. Give the payoff 2-4 seconds where almost nothing moves. Comparisons need reading time.
Split screen or match cut
Split screen puts both states on screen at once — side by side, or stacked in a 9:16 frame. Safest format: both halves are visible simultaneously, so no identity continuity is required. Best for cleaning, room makeovers, car detailing — anything where the environment changes. Write the divider explicitly (“hard vertical split, 50⁄50, thin dark divider line”) or the halves blend into mush.
Match cut keeps one frame and swaps the state on an action: a hand covers the lens, fingers snap, a whip pan, a door slams. Stronger for costume and styling — the reveal is the surprise. It demands a tighter match: write one framing line both states share, subject at the same position and size.
A useful hybrid: the reveal wipe — a hand sweeps across frame trailing the new state behind it. Reads like a match cut, forgives small mismatches; the wipe hides the seam.
Pick one format per video — stacking both in 15 seconds gets you scrolled past.
The seven elements of a before/after
| Element | Transformation version | What to write in the prompt |
|---|---|---|
| Subject | One subject, defined once, in two states | “A woman in her late 20s, curly hair, same person both halves; state A: gray hoodie, slouched; state B: tailored coat, upright” |
| Framing | Identical both states: size, position, height | “Medium shot, centered, head at upper third — identical before and after” |
| Camera | Locked, or one matched move in both halves | “Locked eye-level both halves; or identical 5% push-in in each” |
| Motion | Transition action plus mirrored pose | “Looks down at 2s; hand sweeps the lens at 3s; looks up at 4s in the new state” |
| Light & color | Same light logic; the state carries the shift | “Same soft window light; state A flat and desaturated, state B richer contrast and warmth” |
| Timing | Before 2-3s, transition ~0.5s, after 3-5s, hold 2-3s | “0-3s before, wipe at 3s, after state to 15s, hold still 13-15s” |
| Sound | One transition hit, clean ambience, swell on reveal | “Quiet room tone, one whoosh at the wipe, beat drops on the reveal” |
Framing is the row people skip — and the one that decides whether the comparison reads.
Workflow: steal the transition grammar, not the video
- Find a clip where the cut lands. Not the prettiest one — the one where the reveal gave you a physical reaction. That’s structure.
- Run it through SocialToPrompt. The frame-by-frame breakdown exposes the grammar: matched framing, pose at the cut, the swap moment, the hold after. Videos up to 20s; new accounts get 3 credits, a full run costs 4 (1 extraction + 3 generations).
- Keep the beats, change everything else. Same framing values, same transition timing, same hold. New subject, new state pair. The grammar is the format; someone else’s hoodie is their video.
- Generate halves separately when needed. If a model won’t hold the subject across states, generate both with the identical framing line and composite the transition yourself.
- Check the calendar. 10⁄31 costumes, 11⁄27 gear, 12⁄25 homes, 1⁄1 resolutions. Same grammar, four hooks.
Copy-paste templates
1. Split-screen wardrobe transformation (9:16, ~15s)
Vertical 9:16, split-screen transformation, hard vertical split at center, thin dark divider. Left "before",
right "after", identical framing: a woman in her late 20s, curly hair, medium shot, centered, head at upper
third, plain warm-gray wall. Left: gray hoodie, slouched, flat overcast light, desaturated; she shifts
weight and sighs. Right: same woman, same position and size, camel coat, hair pinned up, upright, richer
contrast and warm saturation; she lifts her chin and smiles. Camera: locked eye-level. Timing: idle motion
0-4s; at 4s both panes snap to a pose with a shutter click; hold still 12-15s. Sound: room tone, shutter
at 4s, faint beat from 5s. No text. Total 15s.
2. Match-cut lens wipe tidy-up (9:16, ~18s)
Vertical 9:16, match-cut transformation. Subject: a cluttered home desk from a seated POV — laptop, mugs,
papers, cables. State A: dull cool light. Framing: POV medium-wide, desk fills the lower two thirds,
identical in both states. Camera: locked POV. Motion: 0-2s a hand enters right and sweeps a cloth across
the lens; as it passes at 2-3s the scene behind is already the new state; hand exits left at 3s. State B:
same desk, same laptop position, now clear — one mug, a small plant, cables coiled, warm directional light.
Timing: dirty 0-2s, wipe 2-3s, clean 3-15s, 2% push-in 3-13s, hold 15-18s. Sound: cloth friction, low
whoosh, quiet tone, light beat from 4s. No text. Total 18s.
3. Top/bottom gear glow-up (9:16, ~20s)
Vertical 9:16, stacked split-screen, thin light divider at mid-frame. Top "before", bottom "after",
identical framing: overhead flat lay of a desk corner on pale wood — keyboard, notebook, one cable, one
mug. Top: worn matte keyboard, tangled cable, scuffed notebook, cool gray light; a hand drops in a second
tangled cable. Bottom: exactly mirrored layout, all aligned — smooth dark keyboard, cable coiled, notebook
squared to the edge, warm key light on a soft gradient; a hand sets down a small desk lamp. Camera: locked
overhead both panes. Timing: both live 0-5s; at 5s a soft click, both freeze 0.5s; bottom animates 5-15s,
top holds; both still 15-20s. Sound: quiet ambience, click at 5s, minimal beat from 6s. No text. Total 20s.
Pre-flight checklist
- [ ] One format per video: split screen, match cut, or reveal wipe
- [ ] Framing written once, declared identical for both states
- [ ] Subject specified as the same subject in both halves
- [ ] Transition named explicitly (wipe, snap, shutter, hard cut)
- [ ] “After” state holds still 2-4 seconds
- [ ] Light logic consistent; the state change carries the color shift
- [ ] One transition cue plus ambience, no audio clutter
- [ ] No real logos, packaging or borrowed likenesses
Common mistakes and FAQ
Q: Why do my two halves look like different people/rooms? The prompt described two scenes, not one subject in two states. Write subject and framing once, add “identical in both states”, and reuse that framing line when generating halves separately.
Q: Dissolve or hard cut? Hard cut on an action, always. Dissolves read as slideshow; the change must land at a moment the viewer feels.
Q: My model can’t keep the subject consistent across states. Now what? Generate two clips with the same framing line and composite the transition in the edit. Or dodge it: split screen shows both states at once and never tests continuity.
Q: Can I use a real creator’s transformation as the “before”? No. Shot grammar is free to learn; someone’s face, footage and captions are not. Keep both states yours and brand packaging out of frame.
Find the before/after clip whose cut gave you a reaction, run it through SocialToPrompt, and rebuild that transition grammar around your own subject tonight.
Share Article