9:16 Vertical AI Video: Which Models Support It, and Sume Auto Limits

Vertical is the default delivery format for short video, but not every model renders it. Some only offer 16:9 and 9:16, some add 1:1 and 21:9, and cropping a landscape clip to vertical throws away most of the frame. Here is what each documents on 2026-10-03.

Support by model

Google's Veo and Omni docs both list 16:9 and 9:16. On Sume, kling-3 takes 16:9, 9:16 and 1:1; the Seedance 2 family lists a wide set from 21:9 down to 9:16, with 4:3, 1:1 and 3:4 in between; and gemini-omni-flash-1.1 takes 16:9 or 9:16.

What sume/auto accepts

Sume's Auto route, model: "sume/auto" on POST /v1/videos, defaults to 720p and 8 seconds and accepts 3 to 10 seconds at 16:9 or 9:16. That covers the common vertical ad. It does not cover 1:1 or 21:9, and it will not run a clip longer than 10 seconds; the docs also say Sume never discloses which family served a request, so pin a model when you need a specific one.

Pass the ratio explicitly. A vertical prompt with no aspect_ratio is not a vertical request.

Delivery notes

A 9:16 clip at 720p is 720 by 1280 pixels, below the 1080 by 1920 most platforms ask for. If the final needs full vertical HD, either pick a model that renders 1080p natively or generate at 720p and upscale. Sume's Timeline renders 1080 by 1920 by default, so a 720p source gets resampled when assembled.

For model-by-model selection beyond ratio, see the five questions to ask before pinning one.