TL;DR
Generate at the platform’s target ratio from the start. Cropping a 16:9 render down to 9:16 cuts heads and removes the action, because the AI composed for landscape. As of June 2026, the four ratios that cover almost everything are 9:16 (TikTok, Reels, Shorts), 4:5 (Instagram feed), 16:9 (YouTube, OTT, web), and 1:1 (carousels, signage). The major AI video tools — Sora 2, Runway Gen-4.5, Kling 3.0, Google Veo 3.1 — all generate 9:16, 16:9, and 1:1 natively, but none output true cinematic 2.35:1, so you render the widest native ratio and crop in post.
The cost of getting the ratio wrong
You spend credits on a clean 16:9 hero shot, then the campaign turns out to need 9:16 for Reels. Cropping the landscape frame to vertical puts the subject’s head out of frame and discards the part of the shot where the action happens. Now you re-generate — and on a tool like Runway, where a 10-second Gen-4.5 clip burns a meaningful chunk of a monthly credit budget, a re-render you could have avoided is pure waste.
The fix is boring but reliable: decide the deliverable ratio before you write the prompt, compose for that ratio in the prompt, and only crop when the source is genuinely wider than the target. This guide gives you the exact numbers per platform and tells you which tool produces each ratio without a workaround.
Platform target ratios and resolutions
Use these as the deliverable spec. Resolutions are the standard upload sizes platforms recommend as of June 2026.
| Platform / placement | Aspect ratio | Resolution | Notes |
|---|---|---|---|
| TikTok, Reels, YouTube Shorts | 9:16 | 1080 × 1920 | Dominant short-form ratio; every major AI tool generates it natively |
| Instagram feed (video) | 4:5 | 1080 × 1350 | Meta’s 2026 feed default; outperforms 1:1 for mobile reach |
| Instagram feed (square fallback) | 1:1 | 1080 × 1080 | Safe choice for carousels and grids |
| YouTube standard, OTT, web hero | 16:9 | 1920 × 1080 (4K: 3840 × 2160) | Default for any horizontal product or brand video |
| Link-preview / share card | 1.91:1 | 1200 × 630 | The Open Graph image ratio; matters for how shares render |
| Cinematic / film look | 2.35:1 or 2.39:1 | crop from 16:9 | No mainstream AI tool generates this natively in 2026 |
| Storefront loop / signage | 1:1 | 2160 × 2160 | Larger square for physical displays |
A near-miss ratio is still a miss. A 1.78:1 (16:9) frame is not a 1.91:1 share card, and platforms will letterbox or crop the difference. Generate the exact target.
Which AI video tools generate each ratio (June 2026)
All four leading tools cover the three core ratios. The differences are in max resolution, clip length, and how cleanly each handles vertical composition.
| Tool | Native ratios | Max resolution | Max clip length | Notes |
|---|---|---|---|---|
| Sora 2 (OpenAI) | 9:16, 16:9, 1:1 | 1080p | short clips | Picker offers 480p / 720p / 1080p; OpenAI has slated Sora 2 for retirement on Sep 24, 2026 |
| Runway Gen-4.5 | 1:1, 4:5, 9:16, 16:9, 21:9 | 720p native, 4K via upscaler | ~40s | Widest native ratio support; upscale to 4K is paid-only |
| Kling 3.0 | 1:1, 16:9, 9:16 | 1080p (4K on premium) | up to ~3 min | Longest clips of the group |
| Google Veo 3.1 | 16:9, 9:16 | 1080p | short clips | 16:9 first; 9:16 added later — no native 1:1 |
Practical takeaways: Runway gives you the most ratio options including native 4:5 and 21:9. Kling is the choice when you need a clip longer than ~40 seconds. Veo has no native square, so generate 1:1 work on Sora 2, Runway, or Kling instead. None of them produces 2.35:1 cinematic natively — render 16:9 (or 21:9 on Runway) with extra headroom and crop.
Safe zones: where platform UI eats your frame
The most common composition mistake is putting the subject where the platform’s own interface covers it. For a 1080 × 1920 TikTok video as of June 2026:
- Usable safe zone: roughly 960 × 1386 px, centered in the frame.
- Top margin: ~140 px (profile bar, device notch / Dynamic Island).
- Right margin: ~164 px (like, comment, share, bookmark icon column).
- Bottom margin: ~324 px for organic posts, and ~370 px for ads because of the CTA button and “Sponsored” label.
Captions make the bottom margin grow. A one-line caption eats ~80 px; a three-line caption with hashtags can eat 250+ px. Keep your subject and any baked-in text inside the central safe zone, and mock up the caption block on top of the rendered frame before you ship.
Compose at generation time, not in the editor
Run this once and you will stop trying to crop your way out of mismatched ratios: take one concept and generate it three times — 9:16, 16:9, and 1:1 — then put them side by side. The same prompt produces meaningfully different compositions in each ratio, because the model repositions the subject and re-balances negative space for the frame. That is the correct behavior. Generating one render and cropping three out of it gives you two versions with clipped heads or an empty half.
When you write the prompt, describe the frame you actually want. For 9:16, ask for the subject framed vertically with headroom and footroom; for 16:9, allow horizontal movement and wider establishing space. Motion has to match the ratio too: a horizontal pan reads naturally in 16:9 and looks cramped in 9:16, and vertical motion is the reverse.
Multi-platform workflow
For a launch that ships to several surfaces, the efficient path is one prompt, several native renders:
- List the target ratios up front (commonly 9:16 + 4:5 or 1:1 + 16:9).
- Write one concept prompt that describes the subject and action, not the framing.
- Generate each ratio natively in your chosen tool — do not crop one master into the others.
- Check the safe zone and composition for each render separately.
- Finalize per platform with the correct captions overlaid, naming files by ratio:
concept_9x16.mp4,concept_4x5.mp4,concept_16x9.mp4.
Three native generations beats nine retakes from a single mismatched master. Generate the 9:16 version first when in doubt — vertical is the hardest to compose for, and a vertical frame crops down to square more gracefully than a landscape frame crops up to vertical.
Common mistakes
- Generating 16:9 and cropping to 9:16. The subject was positioned for landscape; the vertical crop cuts heads and drops the action.
- Ignoring safe zones. TikTok’s caption plus engagement column covers up to ~370 px of an ad frame; LinkedIn auto-trims ratios it dislikes.
- Asking for 2.35:1 cinematic without checking tool support. No mainstream tool generates it natively in 2026 — render the widest native ratio, then crop or letterbox.
- Letting the editor pick the ratio in post. Composition is baked into the frame at generation time; you cannot add it back later.
- Treating 4:5 and 1:1 as interchangeable for Instagram. 4:5 (1080 × 1350) takes noticeably more screen on the mobile feed and is Meta’s 2026 default.
- Choosing Veo for square deliverables. Veo 3.1 has no native 1:1 — use Sora 2, Runway, or Kling for square.
FAQ
- Can I generate cinematic 2.35:1 natively? Not on the mainstream tools as of June 2026. Sora 2, Kling, and Veo top out at 16:9; Runway Gen-4.5 goes to 21:9 (about 2.33:1), which is the closest native option. For true 2.35:1, generate the widest native ratio with extra headroom and crop or letterbox in post.
- Does aspect ratio affect generation quality? Yes. Training data skews toward common ratios, so extreme or unusual ratios sometimes produce odd compositions. Some tools also internally bin to the nearest supported ratio and silently resize, so confirm the output dimensions match what you requested.
- What matters more, resolution or aspect ratio? Ratio drives the viewing experience; resolution drives delivery quality. A 1080 × 1920 9:16 clip beats a 4K 16:9 clip on TikTok every time, because the 16:9 clip gets letterboxed into a small band.
- Which tool should I use for vertical-first work? Any of the four generate 9:16 natively. Runway offers the most ratios including 4:5; Kling handles the longest clips (up to ~3 minutes); Sora 2 and Veo 3.1 are strong for short, photoreal shots. Pick on clip length and ratio coverage.
- What ratio for LinkedIn video? Use 1:1 or 4:5 for the feed and 16:9 for native LinkedIn video. Avoid 9:16 — LinkedIn’s feed crops tall verticals.
- What about VR or 360 video? Different format entirely. General AI video tools do not output equirectangular projection in 2026; use a dedicated 360 tool.