Lyria 3.5 or the free BGM catalog: which bed for your video?
Lyria 3.5 on Sume costs $0.125 a track and is written to your brief. The BGM catalog has no generation charge and is a fixed set of tracks. Which fits?
Use the BGM catalog when a loop that roughly fits is enough: picking costs nothing extra and the track is fixed. Use Lyria 3.5 through Music Router when you need a bed written to your brief: it costs $0.125 per generation, takes a prompt of up to 5,000 characters, and no two runs are identical. For a 15-second ad, the catalog is usually enough.
Side by side
| Question | BGM catalog | Music Router (Lyria 3.5) |
|---|---|---|
| Cost per use | No generation charge | $0.125 per generation |
| Control | Filter by mood, genre, energy, BPM, category | Free-text brief up to 5,000 characters |
| Repeatable | Yes, the same file | No, results are non-deterministic |
| Length | Fixed per track; many loop | Steered in the prompt; no duration field |
| Credit needed | Only the Kevin MacLeod beds (CC BY 4.0) | None; output carries a SynthID watermark per Google |
| Exclusions | Filter | Write them in the prompt; no negative prompt |
Pick the catalog when
- You need the same bed across a batch, so every video shares one sound.
- The video is short and the mood is a common one: corporate, acoustic, beauty, tech, cinematic, kpop, commerce, fashion or lofi.
- You want a loop that is built to repeat under a long video.
- You are prototyping and do not want each draft to cost a generation.
Generate when
A specific story arc or instrument matters, such as a lift at 0:20 or a muted trumpet answer; when the catalog's bgm_pick returns widened or catalog_fallback; or when you want an image to set the mood through image_url. See the match tiers for what those results mean.
Google's Lyria guide states the output is non-deterministic and carries a SynthID watermark. If you need to choose among options, run a few takes and audition them.
A cheap way to decide
Run bgm_pick first. If the tier is strict or relaxed and the sample sounds right, you are done at no cost. If it is widened or catalog_fallback, spend $0.125 on a brief that says exactly what you wanted. Either way, put the bed under the voiceover with a duck_db value so the words stay clear.
The two routes are not exclusive: use the catalog for the bulk of a campaign and generate one bespoke bed for the hero video.
Mix and match
Catalog beds are good for the bulk of a campaign; a generated bed is good for the hero video. Keep the generated brief, since it is a reusable recipe, even though the next run will not be identical. Log which route you used, so the team can see what each video cost in music.
In both cases the Timeline soundtrack object does the mixing: gain, loop, fade out and duck.
Sources
Related posts
More in Comparisons
- Lyria 3.5 vs Suno v6 for ad background music
Lyria 3.5 is an API at $0.08 a song with SynthID. Suno v6 is a subscription with commercial rights only on paid plans. Pick by workflow, then check terms.
- MAI-Transcribe-2 vs Sume STT 1.0: keywords, diarization, price
MAI-Transcribe-2: keyword biasing, diarization, $0.10/hr limited-time. Sume STT 1.0: $0.01 per minute ($0.60/hr), word timings always on, no keyword field.
- MAI streaming regions vs Sume: no region field in the STT request
MAI-Transcribe-2-Streaming lists swedencentral, centralus, southindia (eastus2 soon). Sume's STT request has no region field, so ask support about residency.
- MAI-Voice-2.1 languages vs Sume voice tags: 14 shared, 9 and 2 apart
Microsoft lists 23 TTS languages; Sume's voice library tags 16. Fourteen overlap, nine are MAI-only, two (Japanese, Tagalog) are Sume-only. The full split.
Written by Sume