Vertical 9:16 AI Video Settings for Shorts and Reels

Most short-form video is watched on a phone held upright, so a short drama, a trailer or a hook clip needs to be made in 9:16 from the start. Generating widescreen and cropping later throws away most of the frame and usually cuts a face in half. The good news is that the main AI video models let you set vertical output directly. The details differ per model, and a few settings decide whether the result looks native to vertical feeds or like a cropped afterthought.

This guide lists the vertical 9:16 AI video settings for Shorts, Reels and other vertical feeds on Seedance 2.5, Wan 3.0 and Kling 4.0, all taken from the live model pages, then covers framing, safe zones and export.

The settings, model by model

Model page	How to get 9:16	Resolutions	Duration	Audio
Seedance 2.5 Text-to-Video	aspect_ratio: 9:16 (default 16:9)	480p, 720p, 1080p, 4k	4-30 s	generate_audio, on by default
Seedance 2.5 Image-to-Video	No aspect setting; follows the start image	480p, 720p, 1080p, 4k	4-30 s	generate_audio, on by default
Wan 3.0 Text-to-Video	aspect_ratio: 9:16 (default 16:9)	480p, 720p, 1080p	2-30 s	generate_audio, on by default
Wan 3.0 Image-to-Video	aspect_ratio: 9:16, or leave empty to adapt to the image	480p, 720p, 1080p	2-30 s	generate_audio, on by default
Kling 4.0 Text-to-Video	aspect_ratio: 9:16 (default 16:9)	720p, 1080p	3-30 s	Synchronized stereo audio
Kling 4.0 Image-to-Video	aspect_ratio: 9:16, or leave it out to follow the first frame	720p, 1080p	3-30 s	Synchronized stereo audio

Two things stand out:

1. Text-to-video defaults to 16:9 on all three. If you forget to change it, you get a widescreen clip. Set aspect_ratio to 9:16 explicitly in every request, including API calls.
2. Image-to-video follows the input image on Seedance 2.5, and on Wan 3.0 and Kling 4.0 when you leave the aspect ratio unset. A vertical start frame gives you a vertical video. A landscape start frame gives you a landscape video, no matter what the prompt says.

For model-specific details, our earlier posts answer can Seedance 2.5 generate vertical videos and can Wan 3.0 generate vertical videos.

A copy-paste request

The same vertical shot as an API request for Seedance 2.5 Text-to-Video:

{
  "prompt": "A young florist with a copper bob locks the glass door of her flower shop at night and turns to look at the rainy street. Medium shot at eye level, slow push-in. Neon reflections on wet pavement, soft rain, no music, no subtitles.",
  "aspect_ratio": "9:16",
  "resolution": "720p",
  "duration": 5,
  "generate_audio": true
}

On Wan 3.0 Text-to-Video the same fields work (prompt, aspect_ratio, resolution, duration, generate_audio). On Kling 4.0 Text-to-Video use prompt, aspect_ratio, resolution and duration; audio is always generated.

Which resolution to pick

Resolution is the main cost lever, because all three models price per second of output and scale with resolution. List prices on the model pages when this post was written:

Model	480p	720p	1080p	4k
Seedance 2.5 (text or image to video)	$0.18/s	$0.36/s	$0.90/s	$1.80/s
Wan 3.0 (text or image to video)	$0.05/s	$0.10/s	$0.20/s	n/a
Kling 4.0 (text or image to video)	n/a	$0.157/s	$0.235/s	n/a

The Generate button on each model page shows the current price before you run.

A sensible pattern for vertical episodes:

• Drafts: 480p on Seedance 2.5 or Wan 3.0, 720p on Kling 4.0 (its lowest).
• Final: 720p for most feed content; 1080p when you deliver 1080x1920 or expect viewers on large phones.
• 4k on Seedance 2.5 is rarely worth it for vertical feeds; reserve it for shots you will crop or reuse elsewhere.

Render the final shot at the resolution you deliver. Upscaling a 480p draft is fine for review, not for publishing.

Framing for a tall frame

A 9:16 frame is tall and narrow, so compositions that work in widescreen often fail:

• Put people in the middle third. The top and bottom of a vertical frame are covered by interface elements on most apps.
• Prefer medium shots and close-ups. A wide shot of two people in 9:16 makes both faces tiny. Use wides for establishing shots only.
• Stack, don’t spread. Two characters read better one in front of the other (over-the-shoulder) than side by side.
• Use vertical motion. Stairs, falls, climbing, looking up at a building: movement along the tall axis fills the frame.
• Name the framing in the prompt. “Medium close-up, eye level, subject centered” gives the model a vertical composition to aim for instead of a widescreen layout squeezed into a tall frame.

For image-to-video, compose the start frame with these rules, since the whole shot inherits it.

Safe zones: what the interface hides

Vertical apps overlay the video with the caption, account name and buttons. These overlays sit along the bottom and the right side, and often across the top. Exact sizes vary by app and change over time, so check your own frames rather than relying on fixed numbers.

Before you burn in captions or a title, check a real frame. Pause the clip on a busy moment and export a still (most video players and editors can save the current frame), then open it in any image editor and draw rough boxes over the bottom band, the right-hand button column and the top status area. Even simpler: upload a private test post to the app you publish on and look at it on a phone. Either way you see whether a face, a caption or a key prop would be hidden before the episode goes out.

Rules of thumb that survive across apps:

• Keep captions above the bottom band and away from the right edge.
• Keep faces and key action in the center.
• Put the series title or episode number high enough to clear the top status area. Design title cards at the same 9:16 size as the episode so nothing is rescaled or letterboxed in the edit.

Duration and pacing

All three models generate up to 30 seconds per clip, but vertical drama cuts fast. Shots of 4 to 8 seconds keep the energy up; keep one action per shot. The first two seconds of an episode decide whether a viewer stays, so start on movement or a face, never on an empty establishing shot. The last shot should hold on the hook; see how to write a vertical AI short drama script that ends on a cliffhanger.

Audio for vertical feeds

Many people scroll with the sound off, and many with it on. Plan for both:

• Keep native audio on (generate_audio: true on Seedance 2.5 and Wan 3.0; Kling 4.0 always generates it) for ambience, footsteps and dialogue.
• Write “no subtitles” in the prompt and add captions in the edit, where you control the font, size and position inside the safe zone.
• Write “no music” if you will add a soundtrack later; music baked into each shot makes cutting harder.

Dialogue-heavy shots have their own guide: AI lip sync dialogue scenes with Seedance 2.5.

Export checklist

Before you upload an episode to a vertical feed:

• Every shot is 9:16; none was cropped from widescreen.
• All shots in the episode share one resolution; mixed 720p and 1080p shots are scaled to the delivery size.
• Delivery size matches the source: 1080x1920 from 1080p shots, 720x1280 from 720p shots.
• Captions and titles pass the safe zone check.
• The first frame is a face or movement, not black.
• The last shot holds on the hook for at least three or four seconds.

For the full production process from script to delivery, read how to make an AI short drama with Seedance 2.5, Wan 3.0 and Kling 4.0.

FAQ

What aspect ratio should I use for Shorts and Reels? 9:16. Set aspect_ratio to 9:16 on text-to-video, or use a vertical start image on image-to-video.

Why did my vertical prompt come out widescreen? Text-to-video on Seedance 2.5, Wan 3.0 and Kling 4.0 defaults to 16:9. Writing “vertical” in the prompt does not change the output shape; the aspect_ratio setting does. On image-to-video, check that the start image itself is vertical.

Which resolution is best for vertical AI video? 720p is a good default for feed content and keeps costs down. Use 1080p for final renders you deliver at 1080x1920. Draft at 480p where the model offers it.

Can I convert a 16:9 AI video to 9:16? You can crop, but you lose most of the frame and the composition rarely holds. Regenerating the shot in 9:16 is usually better and, at draft resolution, inexpensive.

How long should each vertical clip be? Four to eight seconds per shot works well for drama. The models allow up to 30 seconds per clip, which is useful for continuous action or long dialogue takes.

Start your next episode in the right shape: open the video generator with Seedance 2.5, Wan 3.0 or Kling 4.0 and set 9:16 before you write the first prompt.