YouTube video A/B testing of three cuts: build them with Sume

YouTube's Made on YouTube 2026 post says creators can use video A/B testing to try up to three different cuts and see which holds attention best. It gives no launch date, so plan now but do not count on it. You can prepare three cuts with Sume by trimming the opening at $0.02 per clip and joining each version in a Timeline 1.0 render at $0.10 per output minute, after an unbilled plan call.

The YouTube claim comes from its Made on YouTube 2026 blog post, read on 2026-10-02, which says the feature is rolling out in YouTube Studio with no timeline. Prices and limits below come from Sume's video trim and Timeline 1.0 docs.

What exactly did YouTube announce about testing cuts?

The blog says YouTube Studio lets creators use video A/B testing to try out up to three different cuts for which holds audience attention best. It does not say which videos qualify, how the winner is picked, how long a test runs or what a cut may change. Today's separate title and thumbnail test is documented elsewhere and picks by watch time, but the post does not say the cut test works the same way.

So treat everything past 'up to three cuts' as unknown. Build your three versions so they differ in one decision each, which is useful whatever the rules turn out to be.

What should the three cuts change?

The most common attention drop is the first seconds, so the cleanest experiment is the opening. Keep the body identical and change only the start. A trim-and-join pipeline suits that, because the body clip is cut once and reused in all three renders.

How do you build the cuts with Sume?

First, cut the pieces with POST /v1/video-trim: one range for the opening, one for the hook line, one for the body. Trim takes a start and exactly one of end or duration, and precision: "exact" is the default frame-accurate re-encode. Source must be at most 1800 seconds, so a long video needs a split first. Each trim is $0.02.

Then plan the joins. POST /v1/timeline-1.0/plan is unbilled and returns duration_seconds, segment_count and estimated_cost_usd_micros without creating a job. Render with POST /v1/timeline-1.0/render once the numbers look right. Timeline needs an audio spine and ordered video[] slots; for a quick cut where the picture's own sound should carry, detach the audio with audio detach and reuse it.

curl -X POST https://api.sume.com/v1/timeline-1.0/plan \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "audio": {
      "url": "https://media.sume.com/artifacts/artf_demo/cut-b-voice.wav",
      "duration_seconds": 52
    },
    "video": [
      { "source_url": "https://media.sume.com/artifacts/artf_demo/result.mp4", "start": 0, "duration": 6 },
      { "source_url": "https://media.sume.com/artifacts/artf_demo/body.mp4", "start": 6, "duration": 46,
        "transition": { "type": "fade", "duration": 0.25 } }
    ]
  }'

What are the sharp edges?

Every URL in a Timeline request must already be a media.sume.com artifact or asset in your workspace, so import first with POST /v1/media-imports. The audio spine sets the output length, and a video slot may trail it by at most 0.5 seconds, so the plan call is the place to catch a gap. Timeline renders 1080x1920 MP4 by default; set output.width and output.height for 16:9, as covered in podcast clip output sizes.

A different audio spine per cut means a different voice track per cut. If the hook line is spoken, you need that audio file; Sume's Timeline does not generate speech inside a render.

Is this worth doing before the feature ships?

Yes in a limited way. Writing the three-cut plan costs nothing, and the plan endpoint is unbilled. Rendering costs cents, not dollars: three one-minute cuts are about $0.30 in Timeline plus a few trims, per the docs, though you should confirm live pricing in GET /v1/catalog.

What Sume cannot do is run the YouTube test, upload the cuts or read the result. It makes the files. Reuse the same approach for paid ads in hook variations, where you can measure outside YouTube today.