Background music has exactly one job: support the thing in front of it without competing. That makes it the easiest music to generate well and the easiest to get wrong, because most people prompt for music they would enjoy listening to rather than music that disappears correctly.
The technical bar is low. The craft bar is a single rule about the mid range.
TL;DR
- Keep the mid range empty. Speech lives there, and this is the rule that decides everything
- Cheapest on 8frame: MiniMax Music v2 at 5 credits, or Lyria 2 at 13 per 30 seconds
- Generate 20 to 30 seconds and loop it, rather than a full-length track
- Free tiers are usually personal use only, which excludes a monetised video
- No transients under words. A snare landing on a syllable erases the syllable
The mid-range rule
Human speech occupies roughly the middle of the frequency spectrum. Any musical element with melodic content there competes directly with your narration, and the listener's attention splits between them.
What this means practically:
- No lead melody in the speech range. Keep melodic content low (bass) or high (bells, pads, texture).
- Pads over tunes. Sustained harmonic texture supports; a tune demands.
- Filter the mids. A high-pass and low-pass carving the middle out of the bed is the single most effective edit you can make, and it takes ten seconds in any editor.
A bed that sounds slightly empty on its own is usually correct under speech. A bed that sounds good alone is usually wrong.
No transients under words
A snare hit, a rim, a plucked note landing on a syllable makes that syllable disappear. For any talking content, prefer beds with soft attacks: pads, sustained strings, filtered synths, ambient texture.
If the bed needs rhythm, keep the transients quiet and let the bass carry the pulse.
What it costs
Canon 8frame prices, a credit is $0.01 at pack rate:
| Route | Credits | Per minute |
|---|---|---|
| MiniMax Music v2 | 5 per generation | depends on length |
| Lyria 2 | 13 per 30 seconds | 26 |
| MiniMax Music 3 | 17 per generation | depends on length |
| ElevenLabs Music | 81 per minute | 81 |
| ElevenLabs SFX | 5 per effect | room tone, texture |
The efficient number: a 30-second loopable bed on Lyria 2 is 13 credits, about thirteen cents, and it covers a ten-minute video. Generating ten minutes directly on ElevenLabs Music would be 810 credits, about $8.10, for something nobody notices loops.
Prompting a bed
- Function first. "Background bed under narration, sparse, no lead melody, nothing in the mid range."
- Tempo and feel. "80 bpm, calm, steady."
- Named instruments. "Warm analogue pad, soft sub bass, occasional high bell." Naming instruments beats naming a genre.
- Dynamics. "Consistent level, no build, no drop." Generated music often has an arc, which is lovely alone and a problem under speech.
- Loopability. Ask for a loop, then test the join ten times and crossfade it in the edit if it does not sit.
The licensing trap
This is where people create problems. Free tiers in AI music are typically personal use only, and a monetised video is commercial use. The distinction catches out a lot of creators who assume "free to generate" means "free to use commercially."
On a paid 8frame plan, output is watermark-free and commercial use is cleared on paid plans. For the current state of third-party free-tier music licensing, including a change worth knowing about, see best free AI music generators.
Where a library still beats generation
Music libraries give you human-performed tracks, clean commercial licences, multiple lengths of the same cue, and stems. Stems in particular are something generation does not provide, and being able to mute the drums under a dialogue section is genuinely useful.
Generation wins on specificity and immediacy: you need a 22-second bed at 80 bpm that gets out of the way, now. That is a frequent, real need, and thirteen cents is the right price.
FAQ
What is the best free background music generator? Any free tier will produce usable beds. The licence is what separates them, not the sound.
How long should a bed be? As short as loops cleanly. 20 to 30 seconds covers most video work.
Why does my music fight the voiceover? Because it has melodic content in the speech range. Carve the mids out.
Is AI background music royalty free? On a paid tier generally yes, subject to terms. Free tiers usually are not.
Empty middle, soft attacks, thirty seconds looped. The 8frame canvas is free and unlimited, and generation is paid from $19/month.