Eleven v4 tags like [light rain] vs a separate sound bed on Sume

ElevenLabs' Eleven v4 announcement says you can direct delivery with inline tags such as [laughs], [said angrily in French accent], [light rain] and [phone buzzing] (read 2026-10-05). The last two are not speech: they ask the model to add ambience or an effect inside the same audio file.

The trade-off

With one file you get a finished take in one call. The cost is that the rain, the buzz and the voice are baked together. If the client wants the rain quieter, or the buzz two seconds later, you regenerate the line. With separate stems you change one number.

The Sume way: voice and bed apart

Sume TTS returns a voiceover only, with no inline effect tags. Ambience or music is a second file that you import or generate, then place in a Timeline render as the soundtrack. The documented soundtrack fields are url, gain_db, loop, fade_out_seconds (up to 10) and duck_db (0 to 20), and ducking needs a real audio spine under it, not silence.

• gain_db sets the bed level.
• loop repeats a short rain loop under a long voiceover.
• fade_out_seconds ends the bed cleanly.
• duck_db lowers the bed while the voice speaks.

Which to pick

A budget example

A 600-character voiceover costs $0.0285 on Sume TTS, billed as 3 cents. Timeline renders are priced separately; check the Timeline docs for the current rate before you plan a batch. The point is that the bed is reusable: pay once for a rain loop and put it under every video.