Prompting11 min read

The 9:16 "Safe Zone" Prompt Kit: Stop Cropping Heads & Captions When You Turn Veo3Gen Clips into Reels/Shorts

A creator-ready 9:16 safe zone prompt kit with a worked rewrite example, a 60-second test, and Veo3Gen-specific workflows to stop cropped heads and covered capt

On this page

TL;DR

If your AI clips look fine in preview but get ruined in Reels/Shorts, it’s often not “bad generation”—it’s bad vertical composition: faces too close to the top edge, no intentional negative space for captions, and critical elements living where platform UI overlays land. Fix it by prompting 9:16 safe-zone framing on purpose (headroom + eyeline + reserved caption space), then doing a fast 60-second safe-zone test on a screenshot before you batch-generate.

Key takeaways

  • Use a 9:16 safe zone prompt: explicitly request headroom, eyeline placement, and “keep lower third clean.”
  • Design negative space (lower third and/or a side column) so captions and UI don’t cover your face or product.
  • Keep the “read” instant: recognizable subject in the first second + one clear motion; don’t rely on subtle movement that disappears on phones (https://www.vidu.com/blog/social-media-video-ai).
  • Keep camera motion limited and contrast high to help clips survive compression (https://www.vidu.com/blog/social-media-video-ai).
  • Generate a single draft first, screenshot it, overlay rough UI/caption bands, revise the prompt, then scale variants.

Why vertical AI videos get “ruined” after export (it’s a phone problem)

Vidu’s social-video testing points to a common trap: AI clips can look interesting on a desktop preview, then fall apart on a phone because the subject is too small or the motion is too subtle to register while scrolling (https://www.vidu.com/blog/social-media-video-ai).

On top of that, platforms enforce their own aspect ratios and file constraints; get even one parameter wrong and you invite automatic cropping or quality loss (https://www.vidu.com/blog/social-media-video-ai). You don’t fix that by memorizing specs—you fix it by generating layout-aware vertical masters that tolerate UI overlays and compression.

What “ruined” usually looks like:

  • Cropped heads (forehead clipped by a top UI band).
  • Captions covering the mouth/product (lower third collision).
  • CTA text unreadable (thin type + busy background + compression).
  • Hook doesn’t read instantly; Vidu notes most AI clips fail the “no second watch required” test (https://www.vidu.com/blog/social-media-video-ai).

The 9:16 safe-zone concept (creator version)

Treat safe zones as priorities, not pixels. Your goal is to generate a layout that can absorb platform overlays.

Safe-zone priorities

  1. Protect the hero: eyes, mouth, hands, product—keep them away from top/bottom edges.
  2. Reserve overlay space: captions (lower third), UI (top/bottom bands), and any edit-layer text you’ll add later.
  3. Make it readable instantly: recognizable subject in the first second, one clear motion/transition, and end before the clip starts to degrade (these were common traits in Vidu’s repeated social-post tests; https://www.vidu.com/blog/social-media-video-ai).

What to protect (non-negotiable)

  • Eyes/forehead (top crop looks accidental immediately)
  • Mouth (caption zone loves to sit where mouths end up)
  • Hands/product (product demos often dip into the lower third)

What to keep empty (on purpose)

  • Lower third: subtitle + hook text zone
  • Upper band: UI + breathing room
  • Optional side column: a clean “caption rail” for CTAs or a logo bug

The 9:16 “Safe Zone” Prompt Kit (copy/paste snippets)

Use these as add-on lines at the end of any scene prompt.

How to combine:

  • Pick 1 framing snippet + 1 negative-space snippet + (optional) 1 text rule.
  • Keep them consistent across variations so your batch outputs frame similarly.

A) Framing snippets (choose one)

  1. Centered talking head (headroom + eyeline)
  • “Vertical 9:16 composition. Head-and-shoulders framing. Subject centered. Eyes on the upper third. Extra headroom above hair. Keep chin fully visible.”
  1. Waist-up demo (caption-safe hands/product)
  • “Vertical 9:16. Waist-up shot. Subject centered. Hold product at chest height. Keep hands and product above the caption area. Stable framing, minimal camera movement.”
  1. Full-body (safe margins)
  • “Vertical 9:16. Full-body shot. Keep subject fully in frame with extra space above head and below feet. Locked camera. Simple background.”
  1. Product close-up (center lock)
  • “Vertical 9:16. Tight product close-up. Product centered with safe margins. Smooth single rotation. Minimal motion besides the product.”

Vidu reports a rotating-product prompt produced centered, smooth rotation in some tests but also showed mid-clip jitter in others—so keep motion simple and inspect early (https://www.vidu.com/blog/social-media-video-ai).

B) Negative-space snippets (choose one)

  1. Lower-third reserved for captions (most common)
  • “Leave generous clean negative space in the lower third for captions; keep face, mouth, hands, and product out of that area.”
  1. Top-negative-space hook
  • “Leave large clean negative space at the top for hook text; keep subject lower-middle; avoid top-edge cropping.”
  1. Side ‘caption column’
  • “Leave a clean vertical column on the right for captions/CTA; place subject on the left third; keep background simple and high contrast.”
  1. All-sides safe margins
  • “Safe margins on all sides; no important elements near edges.”

C) Text rules (use sparingly)

  1. If you must bake text in
  • “If any on-screen text appears, keep it large, bold, high-contrast, and away from edges; place it only inside the reserved negative space.”
  1. If you want zero text conflicts
  • “No embedded text, no subtitles, no logos.”

D) Quick “avoid” line (add when outputs are too tight)

  • “Avoid extreme close-up; avoid cropping head; avoid subject drifting off-center; avoid important details near edges.”

Worked example (before/after): fixing the Reel that crops the head

This is the exact rewrite pattern most creators need.

Before (tight crop + caption collision)

“A creator talks to camera in a bright kitchen, holding a skincare bottle, energetic, modern, high quality, realistic.”

Why it fails:

  • No instruction for headroom → face often rides high.
  • No reserved lower third → hands/product drift into caption space.
  • Variants frame differently because layout wasn’t specified.

After (a layout-aware 9:16 safe zone prompt)

Copy/paste and customize:

“Vertical 9:16 UGC-style video in a bright kitchen. Waist-up shot of a creator holding a skincare bottle at chest height. Subject centered. Eyes on the upper third. Extra headroom above hair. Leave generous clean negative space in the lower third for captions; keep mouth, hands, and product above the caption area. Stable framing, minimal camera movement. High contrast between subject and background. One clear motion: raise the bottle slightly and point to it.”

What this fixes (specifically):

  • Headroom protects the forehead from top UI.
  • Lower-third negative space prevents captions from covering mouth/product.
  • One clear motion matches what Vidu saw in repeatable social-clip wins (recognizable fast + one motion) (https://www.vidu.com/blog/social-media-video-ai).

Make it faster inside Veo3Gen (when you’re ready to scale)

Veo3Gen supports 9:16 output (and 16:9) and supports text-to-video and image-to-video, plus first-and-last-frame control on Veo 3.1. It also generates native synchronized audio (dialogue, SFX, music) in a single pass—so your Reel draft can come out with audio without a separate step.

When your prompt is stable and you’re testing multiple hooks, backgrounds, or product angles, use Veo3Gen’s developer API to generate variants programmatically. CTA: If you’re doing structured variant testing, start with a single safe-zone master and then automate batches via /api.

Two generation recipes (pick one on purpose)

Recipe A: Vertical-first master (best for Reels/Shorts)

Use this when most views are on phones.

Recipe B: One master that repurposes to 16:9 (use with caution)

Use this when you must deliver both vertical and widescreen from the same concept.

  • Compose for a center-safe column (imagine 9:16 inside 16:9).
  • Keep the hero centered; avoid edge storytelling.
  • Avoid embedded text; add text per format in the edit.

Trade-off: the more you future-proof for widescreen, the smaller your subject becomes—which matches Vidu’s phone-scale failure mode (https://www.vidu.com/blog/social-media-video-ai).

The 60-second safe-zone test (do this before batch generation)

This prevents wasting generations on the wrong framing.

  1. Generate one draft (9:16) with your best prompt.
  2. Screenshot a representative frame (usually 0–1s when the hook must read).
  3. Draw three rough bands on the screenshot:
    • top UI band
    • bottom UI band
    • a thicker “caption block” where subtitles typically sit
  4. Check collisions: do the bands overlap eyes, mouth, product, or hands?
  5. Revise prompt using one framing snippet + one negative-space snippet.
  6. Only then batch-generate (new hooks, outfits, backgrounds).

This aligns with Vidu’s guidance that a good social clip should be intentional on a phone screen and not require a second watch to understand (https://www.vidu.com/blog/social-media-video-ai).

Common failure modes (and the exact prompt fix)

Failure: forehead/chin clipped

Fix lines to add:

  • “Extra headroom above hair. Keep chin fully visible. Safe margins on all sides. Avoid extreme close-up.”

Failure: captions cover mouth/product

Fix lines to add:

  • “Leave generous clean negative space in the lower third for captions; keep mouth, hands, and product above the caption area. Hold product at chest height.”

Failure: subject drifts off-center mid-clip

Fix lines to add:

  • “Stable framing, minimal camera movement. Avoid subject drifting off-center.”

If you have a strong reference frame, consider image-to-video: Vidu observed reduced drift when pinning the first frame with an uploaded image (https://www.vidu.com/blog/social-media-video-ai). (Veo3Gen supports image-to-video and first-and-last-frame control on Veo 3.1.)

Failure: motion doesn’t register while scrolling

Fix lines to add:

  • “One clear motion: (turn / pour / point / reveal). High contrast subject/background.”

Vidu lists close framing, limited camera movement, and high contrast as compression-survival traits (https://www.vidu.com/blog/social-media-video-ai).

Failure: clip degrades near the end

Fix:

Checklist

FAQ

How do I stop Reels from cropping the top of the head in 9:16?

Add “eyes on the upper third” + “extra headroom above hair” + “safe margins on all sides,” then validate with a screenshot overlay before you generate variants.

How do I leave space for captions without making the shot feel empty?

Reserve a clean lower third, but keep the subject large (waist-up or close portrait) and include one clear motion so the frame still feels active while scrolling (https://www.vidu.com/blog/social-media-video-ai).

How do I repurpose a 16:9 AI video to 9:16 without losing faces?

Compose for a centered “safe column” from the start: keep faces and key objects centered, avoid edge storytelling, and don’t embed text near the sides.

How do I keep the subject from drifting off-center during the clip?

Prompt “stable framing, minimal camera movement.” If you can, use image-to-video with a strong first frame—Vidu observed reduced drift when pinning the first frame with an uploaded image (https://www.vidu.com/blog/social-media-video-ai).

Should I bake on-screen text into the generation?

Only when the text is non-critical or scene-native and you’ve reserved clean, high-contrast negative space. Otherwise keep the generation clean and add text per platform in editing.

Use this at the end of any prompt and replace the brackets:

“Vertical 9:16 composition. [SHOT TYPE: head-and-shoulders / waist-up / full-body / product close-up]. Subject clearly recognizable within the first second. Subject [centered / left third / right third]. Eyes on the upper third. Extra headroom above hair. Leave generous clean negative space in the lower third for captions; keep mouth/hands/product out of the caption area. High contrast between subject and background. Stable framing, minimal camera movement. One clear motion: [single action]. Avoid extreme close-up; avoid cropping head; avoid important elements near edges; avoid subject drifting off-center.”

Generate vertical-safe variants faster with Veo3Gen

Once your safe-zone layout is consistent, scaling becomes a variants problem (new hooks, products, backgrounds)—not a framing problem. Veo3Gen provides an affordable way to access Google’s Veo 3.1 video models without Google’s enterprise pricing, with three modes (Veo 3.1 Fast, Quality, Lite), plus 9:16 support and native synchronized audio in one pass.

CTA: If you want to iterate without “use it or lose it” pressure, Veo3Gen offers pay-as-you-go credits plus optional monthly plans, and purchased credits do not expire. Check /pricing and use the free credits for your first safe-zone test run.

Start creating with Veo3Gen

Veo3Gen gives you affordable Veo 3.1 video generation with native audio, up to 4K, and credits that never expire — with free credits to start.

Sources

Limited Time Offer

Try Veo 3 & Veo 3 API for Free

Experience cinematic AI video generation at the industry's lowest price point. No credit card required to start.

Get Started FreeView Cheapest Pricing
← Back to all posts