A 10-minute TikTok ad bed: six Lyria tracks, one concat, $1.76
TikTok non-Spark ads run up to 10 minutes. Lyria makes tracks of a couple of minutes, so six beds plus one concat job and a $1.00 render cover 600 seconds.

To put one continuous music bed under a 10-minute TikTok ad, generate six tracks of about two minutes, join them with one Timeline audio concat job, and render the ad as 600 seconds. On Sume the sound side costs 6 x $0.125 + $0.01 = $0.76, and the render is 10 minutes x $0.10 = $1.00, so $1.76 before footage and voice.
TikTok's ad specification says non-Spark video ads can run up to 10 minutes, with a file of at most 500 MB, in .mp4, .mov, .mpeg, .3gp or .avi, and a bitrate of at least 516 kbps (read 2026-10-08). Google's Lyria 3.5 page describes songs of "a couple of minutes", controllable in the prompt (read 2026-10-08). The gap between those two lengths is what the concat job fills.
Why six tracks, not five
600 seconds divided by a 2-minute track is five tracks, but the Music docs say the prompt steers length and no duration field exists, so a 2-minute request can return a bit short. Buying one extra track ($0.125) is cheaper than a failed render ($1.00). The concat job then joins up to 20 parts, so six is well inside the limit.
| Line | Quantity | Arithmetic | Cost |
|---|---|---|---|
| Music generations | 6 | 6 x $0.125 | $0.75 |
| Timeline audio concat | 1 job | 1 x $0.01 | $0.01 |
| Timeline render | 10 output minutes | 10 x $0.10 | $1.00 |
| Total, excluding footage and voice | $0.75 + $0.01 + $1.00 | $1.76 |
Joining the beds
The Timeline audio concat takes parts[] of 1 to 20 Sume-hosted files, joins them in the sample domain with no silence at the seams, and the output is at most 1800 seconds. Keep the default wav output: the docs say mp3 adds priming padding at every edge, and five joins would mean five padded edges. A wav of 600 seconds is large. If the output keeps 44.1 kHz stereo at 16-bit (176,400 bytes per second), it is about 106 MB; the docs do not state the output sample rate, so check the real file.
Each track should carry the same energy at its ends, or the join will be audible even if it is sample-exact. In the prompt, ask for a quiet first and last bar: begin and end on a sustained pad. The Music 1.0 page notes that section markers such as [0:00-0:30] Intro: shape each track.
The 600-second edge
Set audio.duration_seconds to 600 on the render. A 601-second ad bills 11 minutes ($1.10) and misses TikTok's 10-minute limit, and the render reserve is ceil(audio.duration_seconds / 60) minutes. The spine can be longer than the output because duration_seconds sets the output length, so a six-track spine of about 12 minutes is trimmed to 10.
A 10-minute ad is a long video to carry one bed: consider changing the brief every second track (a different lead instrument at least) so the sound moves, and check that the finished file stays at or above 516 kbps, which TikTok lists as the minimum bitrate.
Sources
Related posts
More in Use cases
- 10 two-second ad hooks on Wan 3.0: $2.50 at 720p on Sume
Wan 3.0 is the only Sume video model that accepts 2-second clips. Ten hooks cost $1.30 at 480p, $2.50 at 720p or $5.00 at 1080p; others start at 3 to 5 seconds.
- 100-second AI presenter explainer: two 50-second jobs, $24.50 at Plus
A 100-second presenter video on Sume is two 50-second avatar jobs plus a Timeline join: $18.60 Standard, $24.70 Plus, $55.20 Max with the $0.20 join.
- A mixed 1,000-prompt library: ads, demos and teasers by model, priced
Matching the model to the clip type puts a 600/250/150 library at $1,162.50 on Sume, against $3,722.50 if every clip ran on seedance-2-fast. The arithmetic.
- 104 newsletter headers in 4:1: $7.80 at 0.5K on Nano Banana 2.1
A year of twice-weekly newsletters needs 104 wide headers: $7.80 at 0.5K or $10.40 at 1K on Nano Banana 2.1 on Sume, the GPT 2.5 3:1 cap aside.
Written by Sume