Cut video to the beat: Sume has no beat finder, so set the starts
YouTube's audio tools include find beats. Sume does not detect beats; for a fixed tempo you compute slot start times from BPM and send them to Timeline.

YouTube's audio page lists a find-beats tool; Sume has no beat detector in its docs. If you know the track's tempo, you can still cut to the beat: one beat is 60 divided by the BPM in seconds, so at 120 BPM a bar of four beats is 2 seconds. Put each video slot's start on a multiple of that number and Timeline will cut there.
YouTube's tool list is from Add & edit audio, read 2026-10-03. The slot rules are in the Timeline 1.0 doc.
How do I compute the starts?
Pick the BPM, beats per cut and the track length. Starts are n * beats_per_cut * 60 / bpm. A slot must be at least 0.2 seconds, video[0].start must be 0 and starts must increase, so a very fast tempo with one beat per cut can fail the 0.2-second floor only above 300 BPM; at normal tempos it is fine.
import json
BPM = 120
BEATS_PER_CUT = 4
TRACK_SECONDS = 24
CLIPS = [f"https://media.sume.com/artifacts/artf_demo/c{i}.mp4" for i in range(1, 4)]
cut = BEATS_PER_CUT * 60 / BPM
slots = []
t = 0.0
i = 0
while t < TRACK_SECONDS:
d = min(cut, TRACK_SECONDS - t)
slots.append({"source_url": CLIPS[i % len(CLIPS)], "start": round(t, 3), "duration": round(d, 3)})
t += cut
i += 1
print(json.dumps({"video": slots}, indent=2))What about music with no known tempo?
The Music Router and Music 1.0 have no tempo parameter in their docs, but you can ask for a tempo in the prompt, for example "120 BPM". Treat the result as approximate: generated audio may drift from a perfect grid, so check the first and last cuts by ear before you commit the batch. Music generation is $0.125 per generation.
How tight is a cut?
Cuts land on the frame grid of the output. Default output fps is not fixed unless you set 24, 25, 30 or 60, so set fps once for a series. At 30 fps a frame is 0.033 seconds.
| BPM | Seconds per beat | Seconds per 4-beat cut |
|---|---|---|
| 90 | 0.667 | 2.667 |
| 100 | 0.6 | 2.4 |
| 120 | 0.5 | 2.0 |
| 140 | 0.429 | 1.714 |
What Sume does not do
No beat analysis, no waveform view and no auto-cut to music. The tempo has to come from you.
Sources
Related posts
More in Use cases
- Day-in-my-life Reel from 12 clips: one Timeline render and its cost
Cut 12 phone clips into a 48-second day-in-my-life Reel: slot timings, a silent spine with a music bed, and what the render costs at Sume's listed rates.
- Declutter a room photo: mask out furniture on ChatGPT Image 2.5
Remove furniture from a listing photo with a mask_url edit on ChatGPT Image 2.5: $0.065875 per attempt on Sume, so 20 rooms with two attempts each cost $2.635.
- Demand Gen A/B experiment: 50 conversions per arm, then what?
Google asks for at least 50 conversions per arm in Demand Gen conversion experiments. What that means for how many AI video variants you can afford to test.
- Dialogue-only subtitles vs CC: Netflix's two English tracks, burned
Netflix offers English (dialogue only) and English (CC) as separate options. How the two differ, and how to burn either version of one clip with Sume cues.
Written by Sume