Avatar video for a 4:5 feed: render 3:4, then crop to 0.9375 height

Sume Avatar 1.0 does not offer a 4:5 aspect ratio. Render at 3:4, then crop the height to 0.9375 of the frame with Sume's video filter, which costs $0.02 per encode and has a free check endpoint. Add captions after the crop, not inside the avatar job, so the crop cannot cut them.

The Avatar Video docs (read 2026-10-06) list five ratios: 1:1, 3:4, 9:16, 4:3 and 16:9, with 9:16 as the default. Feed placements that favor 4:5 sit between 3:4 and 1:1, and 3:4 is the closest taller option.

The arithmetic

A 3:4 frame is 3 wide by 4 tall. A 4:5 frame is 4 by 5, so it is a little wider for its height. Keep the full width and shorten the height: 3 divided by 0.8 is 3.75, and 3.75 divided by 4 is 0.9375. You lose 6.25 percent of the height, or 3.125 percent at the top and bottom if you center the crop.

In the video filter's crop op, x, y, width and height are fractions of the source frame. Width must be between 0.05 and 1, and x plus width, and y plus height, must not exceed 1. The crop that centers the 4:5 window is x 0, y 0.03125, width 1, height 0.9375.

Check first, caption last

Run the same program against POST /v1/video-filter/check first. It is unbilled, returns diagnostics and the compiled filters instead of a 400, and does not create a job. Then submit the encode for $0.02. The source must already be a media.sume.com URL, which a finished avatar video is.

Do the captions after the crop. Inline captions are burned into the final MP4 by the avatar job, so a crop that trims 3 percent off each edge can touch a caption placed near the bottom. A standalone caption job on the cropped file is $0.20 and lets you move the line with the placement setting. The honest tradeoff: that is $0.22 on top of the render, and two extra steps.

When not to bother

If the feed accepts 3:4 or 1:1 natively, skip the crop and use the ratio the avatar job already supports. If a face sits near the top or bottom edge of the frame, ask for a scene prompt that leaves headroom, because the crop does not know where the face is.