ImageKit video smart crop fo-face vs a Sume 9:16 reframe

ImageKit can reframe video to a new aspect ratio and keep a face centred with its fo-face smart-crop option, which tracks the face through the clip. Sume does not: its reframe tools are a fixed crop rectangle, a fit mode that scales the whole frame, and per-shot crops you choose yourself.

ImageKit's rules are from its Video resize and crop page, Sume's from the video filter and Timeline 1.0 docs, read on 2026-10-03.

How does ImageKit's smart crop work?

The aspect-ratio parameter is ar-<width>_<height>, and it must be used with either a height or a width, not both. The default crop strategy is c-maintain_ratio. Smart crop adds a focus option: fo-auto detects and tracks the most important objects, fo-face follows faces, and fo-<object> follows a supported object.

There are caveats. Smart crop only works with c-maintain_ratio and cm-extract, it must be the first transformation step, and it cannot be combined with rotation in the same step.

What does Sume offer for 16:9 to 9:16?

There are three documented pieces, and none of them follows a subject.

Video filter ops[] has a crop op: a rectangle as fractions of the source frame, x and y between 0 and 1, width and height between 0.05 and 1, with x+width and y+height at most 1. The compiler even-rounds for yuv420p. The same crop applies to the whole clip, so one crop cannot follow a moving face.

Timeline 1.0 renders 1080 by 1920 by default, and each video[] slot has a fit of cover (the default, which fills the frame and crops the centre), contain, stretch or blur. cover is a centre crop, which is fine for a centred speaker.

Video trim cuts a range into its own clip, so you can crop each shot separately.

How do you follow a speaker by hand on Sume?

Split by shot and crop each shot where the speaker is. Trim each shot into its own MP4, run video filter with a crop whose x points at the speaker, then place the filtered clips in order in video[]. Because the crop already has a 9:16 shape, the slot fit cover scales it with little further loss.

Before spending anything, POST /v1/video-filter/check validates the program and returns next_action with no charge and no job. That makes it cheap to try several x values.

What does the manual route cost in time and money?

For a 10-shot clip you pay for 10 trims and 10 filters at $0.02 each, plus a Timeline 1.0 render at $0.10 per output minute rounded up. Those are the public rates from the docs; confirm them in GET /v1/catalog. The free POST /v1/video-filter/check lets you test every crop before any of it is billed.

The larger cost is your attention: someone picks the x for each shot. For a talking-head video with two or three speaker positions that takes minutes. For a sports clip with a moving subject it does not scale, and that is where tracking tools earn their keep.

When is ImageKit the better choice?

Pick ImageKit when subjects move a lot inside a shot, such as a handheld walk-and-talk, and automated tracking is worth the occasional miss. Pick Sume when the speaker is mostly still, when you want a crop you can inspect and repeat, or when the clip came out of Sume and the next step is a timeline render or an ad upload.

Whatever you pick, check a few frames before publishing. With Sume, video frames pulls stills from the cropped clip so you can see what the framing really did.