Test three Reel hooks from one clip: three video trims plus renders

Three hook variants of one Reel cost $0.36 in Sume fees if each Reel is 60 seconds or less: three video trims at $0.02 each ($0.06) plus three Timeline renders at $0.10 each ($0.30). The trick is to cut each hook with video trim, then place it as slot one in front of a shared body, so only the first seconds differ.

The test design

Instagram's Reels guide (read 2026-10-03) calls 15 to 60 seconds the sweet spot for tutorials, day-in-my-life clips and most creative formats, so keep the Reel under 60 seconds and each Timeline render stays in its first billed minute. The metrics come from Instagram after posting; Sume does not read them.

Trim each hook

Use precision: "exact" (the default) so the cut is frame-accurate. A hook of 3 seconds is well above the 0.2 second minimum. In each render, slot one is the trimmed hook with source_in 0, followed by the same body slots shifted by 3 seconds.

def reel_slots(hook_url, body_url, hook_s=3, body_s=42):
    return [
        {"source_url": hook_url, "start": 0, "duration": hook_s},
        {"source_url": body_url, "start": hook_s, "duration": body_s},
    ]

hooks = ["h1", "h2", "h3"]
for name in hooks:
    slots = reel_slots(
        "https://media.sume.com/artifacts/artf_%s/hook.mp4" % name,
        "https://media.sume.com/artifacts/artf_demo/body.mp4",
    )
    print(name, sum(s["duration"] for s in slots))

Keep the audio spine the same

Timeline wants one spine length for every variant. If the hooks have different voice lines, generate or trim them to the same duration, or let audio.duration_seconds follow each variant and accept that a few variants may differ by a second. Use the unbilled plan endpoint to confirm billable_minutes is 1 for each before you render.

What Sume does not do

Sume does not run the experiment or tell you which hook won. It makes the variants cheaply and consistently. To look at what viewers saw in the first seconds, sample stills with video inspect and compare them yourself.