Native 1:1 AI video for a TikTok ad, or 9:16 plus a $0.02 crop?

The constraint

TikTok's non-Spark ad specification (read 2026-10-09) lists 1:1 videos at a minimum of 640x640 pixels, next to 9:16 at 540x960 and 16:9 at 960x540. Sume's Gemini Omni Flash 1.1 accepts 3 to 10 seconds at 360p, 720p, 1080p or 4K in 16:9 or 9:16 only, per the video docs, so it cannot return a square clip. You can either choose a model whose supported_aspect_ratios includes 1:1, or generate vertical and crop.

Option A: a model that lists 1:1

The Sume video docs show an example GET /v1/videos/models row for seedance-2 with ratios 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16 and durations of 4 to 15 seconds. Always read the live list before you submit, since each model reports its own supported_aspect_ratios. At the list prices read on 2026-10-08, a 6-second 1:1 clip is 6 x $0.17577 = $1.05462 at 480p and 6 x $0.378 = $2.268 at 720p.

Option B: vertical, then crop

Video filter takes a crop op as fractions of the source frame. For a centered square from a 9:16 clip, width is 1, height is 0.5625 and y is (1 - 0.5625) / 2 = 0.21875. One job is $0.02 and the check route is free. A 6-second Wan 3.0 clip at 720p is 6 x $0.125 = $0.75, so the vertical-plus-crop route is $0.77 against $2.268 for a seedance-2 square at the same tier.

The resolution trap

Cropping a square from a vertical frame shrinks it to the frame's width. A vertical clip that is 480 pixels wide gives a 480x480 square, below TikTok's 640x640 minimum, while a 720-pixel-wide clip gives 720x720. Probe the first output with video inspect before you pay for a batch, since the docs do not give pixel widths per tier. If the width is under 640, generate at the next tier.

Other square uses

The same crop works for any placement that wants a square. YouTube's Shorts page lists square as an accepted shape for videos up to three minutes, so the 1:1 file is also a candidate for a Short. Check the list for each model before you rely on it: the video docs say each model reports its own supported ratios, durations and resolutions, and that limits differ.

If you need the square to follow a face, a fixed center crop may cut it off. The y fraction is yours to set, so move it from 0.21875 toward the top when the subject sits high in the frame, and run the free /v1/video-filter/check call first to confirm the program is valid before you pay for the encode.