YouTube Shorts retention is decided in the first second. Not the first three — the first one. If the opening frame does not promise curiosity, contrast or a clear payoff, viewers swipe before the main beat even loads. The 10 prompts below are first-second openers: each stages a single visual hook, leaves caption room at the top, and stays short enough that the algorithm can chain into your body content. They are written to render cleanly on the AI video models creators actually use as of June 2026 — Google Veo 3.1, OpenAI Sora 2 and Kuaishou Kling 2.5 — and they assume you composite the on-screen text in post, not in the model.
TL;DR
- Hooks belong in the first 1 second; keep the generated clip to 2–4 seconds and edit it into a longer Short (YouTube allows up to 3 minutes / 180s, but the data sweet spot is 15–30s).
- Always render 9:16 at 1080×1920. Square or landscape uploads show with safe-area cropping and lose feed real estate.
- Five prompt levers do the work: lens, light state, one sub-second motion, high-contrast palette, and short duration.
- Do not let the model render the caption text — it warps. Leave top room and add type in CapCut, Premiere or Descript.
- Model fit (June 2026): Veo 3.1 for photoreal faces and synced micro-expressions, Sora 2 for stylized first-frame contrast, Kling 2.5 Turbo for cheap fast iteration on a single take.
What a high-quality hook prompt contains
Five required elements:
- Lens: natural 50mm or talk-to-camera 35mm; never anamorphic flare on a Shorts hook
- Light state: soft window daylight, ring-light flat, or warm bedroom lamp; never moody dark
- Motion: one explicit action under 1 second — arrow tap, zoom, shock-face, countdown swipe
- Palette: high-contrast color clash; never muted
- Duration restraint: Shorts hooks live in 2–4 seconds; never longer than 6
Which model for which hook (June 2026)
| Model | Best for | Native clip length | Native audio | Notes |
|---|---|---|---|---|
| Veo 3.1 (Gemini app, Flow) | Photoreal talk-to-camera, synced lips and micro-expressions | up to ~8s | Yes (dialogue + SFX) | 1080p; reachable on Google AI Pro $19.99/mo; Ingredients-to-Video keeps a face consistent across a series |
| Sora 2 | Stylized first-frame contrast, surreal POV opens | short clips | Yes | Strong on bold color and motion; best when the hook is a look, not a line |
| Kling 2.5 Turbo | Cheap, fast iteration on one continuous take | 5s or 10s | Lip-sync on newer builds | Topped the Artificial Analysis leaderboard; lowest cost-per-try, ideal for A/B testing hooks |
For talk-to-camera authority hooks where the mouth has to match a line, prefer Veo 3.1. For a striking still-into-motion hook with no spoken line, Sora 2 or Kling are cheaper to iterate.
10 copy-ready prompt templates
1. “Watch til the end” arrow
Best for: Creator opener, listicle Shorts
Vertical 9:16, a young creator looks at camera in soft window daylight pointing toward an empty top-right area with a clear hand gesture as if a caption arrow lives there, plain pastel background, sharp focus on face and hand, room at top for caption, 3 seconds2. 3-second curiosity-gap open
Best for: Educational Shorts, story opener
Vertical 9:16, a young creator leans toward camera with a slightly raised eyebrow and a "wait until you see this" expression, soft ring light, plain warm wall behind, sharp focus on the eyes, room at top for caption, 3 seconds3. Before-and-after split first-frame
Best for: Transformation Shorts, fitness, design
Vertical 9:16, a clean side-by-side split frame holds a "before" room on the left and an "after" room on the right, slow tiny zoom in on the right side, soft daylight in both halves, plain neutral palette, room at top for caption, 4 seconds4. “Tell me without telling me” reveal
Best for: Niche-identity Shorts, comedy hook
Vertical 9:16, a young person in soft daylight gestures at an object beside them with a knowing smirk as if saying "you know what this means", plain warm wall behind, sharp focus on the face and object, room at top for caption, 3 seconds5. Direct talk-to-camera scroll-stop
Best for: Creator authority hook, opinion Shorts
Vertical 9:16, a creator looks directly into camera with calm confident expression and opens their mouth as if starting a sentence, soft ring light, plain pastel background, sharp focus on the eyes, room at top for caption, 3 seconds6. On-screen big text question
Best for: Question-led Shorts, polls
Vertical 9:16, a creator points up and to the left as if a large question caption lives there, soft window daylight, plain warm wall behind, sharp focus on the face and hand, ample empty area at top for caption, 3 seconds7. Shock-face reaction freeze
Best for: Reaction Shorts, story payoff
Vertical 9:16 close-up of a young creator going from neutral to wide-eyed open-mouth shock in under 1 second, soft daylight, plain background, sharp focus, room at top for caption, 3 seconds8. Countdown 3-2-1 open
Best for: Listicle, tutorial hook
Vertical 9:16, a young creator holds up three fingers facing camera and slowly drops to two and one in a smooth single beat, soft window daylight, plain pastel background, sharp focus on the hand and face, room at top for caption, 4 seconds9. “POV: you just” scene
Best for: Relatable Shorts, comedy
Vertical 9:16, a first-person view from the doorway of a small messy kitchen as if you just walked in, slow forward step, soft warm interior light, plain frame, room at top for caption, 4 seconds10. “No one talks about” hook
Best for: Insight Shorts, contrarian opinion
Vertical 9:16, a creator looks at camera with a slightly conspiratorial expression and gestures toward an empty space beside them, soft warm interior light, plain wall behind, sharp focus on the face, room at top for caption, 3 secondsCommon mistakes
- Hook longer than 4 seconds — viewers swipe before the payoff lands
- Muted palette — Shorts need contrast to pop against a fast-scrolling feed
- Letting the model render the on-screen caption — text morphs frame to frame; composite it in CapCut, Premiere or Descript at 1080×1920
- Two facial expressions in one second — pick one shift, not a sequence
- Camera shake or anamorphic flare — reads “film”, not “Shorts”
- Uploading anything but 9:16 — square and landscape clips get safe-area cropped and lose feed space
How to push results further
- Pair templates 1 (arrow) and 6 (question) with a real caption added in post — the model holds the room, you add the text
- Templates 3 (split) and 7 (shock) work best when the “after” or “shock” frame is visibly different in color from the first frame
- For authority Shorts (template 5), keep the creator centered with the eye line dead at camera; wandering eyes break the scroll-stop, so prefer Veo 3.1 here for lip and gaze fidelity
- Series strategy: lock background + lighting + lens and swap only the creator’s action so the channel reads as one show. On Veo 3.1, feed the same reference image through Ingredients-to-Video to keep the face consistent across episodes
- Iterate cheap, finish good: A/B several hook takes on Kling 2.5 Turbo (5s, low cost per try), then re-render the winner on Veo 3.1 for the final upload
- 3 seconds is the safest length; only push to 4 when the second beat (a zoom or countdown drop) is written into the prompt
FAQ
Q: How long should the generated clip be vs. the Short itself?
A: Keep the AI-generated hook clip to 2–4 seconds and edit it onto the front of your full Short. YouTube allows Shorts up to 3 minutes (180s), but the engagement sweet spot is 15–30 seconds; the hook just has to survive the first second.
Q: Can AI generate the on-screen caption text?
A: Not reliably on Veo 3.1, Sora 2 or Kling — text drifts and warps across frames. Generate the empty top room and add captions in CapCut, Premiere or Descript so type stays crisp at 1080×1920.
Q: How do I know if a hook is working?
A: Watch the YouTube Shorts retention curve at the 3-second mark in Studio. If the drop is steep, the hook failed; iterate on the first second only, not the rest of the video.
Q: Best resolution and aspect ratio?
A: 9:16 at 1080×1920, always. Square (1:1) and landscape uploads are shown with safe-area cropping in the feed and visibly lose space.
Q: Which model handles talk-to-camera best as of June 2026?
A: Veo 3.1 for photoreal faces and synced micro-expressions (reachable on Google AI Pro at $19.99/mo); Sora 2 for stylized first-frame contrast; Kling 2.5 Turbo when you need a longer single take with the cheapest iteration cost.
Q: Do I need a paid plan to make these?
A: There is a free path on each (Gemini/Google Labs free tier for Veo, free credits on Kling), but free quotas are tight for the volume hook-testing needs. Google AI Pro at $19.99/mo gives reliable Veo 3.1 access; Kling’s paid tiers start around $6.99/mo as of June 2026.
Related articles
- Social Ad Hook Video Prompts
- Viral Shorts Video Prompts
- Cinematic AI Video Prompt Examples
- YouTube Thumbnail Prompts
- Social Ad Hooks sub-hub
- All AI Video Prompts
Tags: #social-ad #youtube-shorts #Hook #Video generation #Prompt