Skip to content

Seed Audio 1.0

ByteDance's sound-effect model, tagged natural: describe a noise and it makes one that sounds like the real thing, with the length chosen by the model.

Updated November 9, 2026

A humpback whale gliding through golden clouds above a 1950s seaside town at dusk

Example image. Write what you want to see and press Create. The image appears in the same frame within seconds. Video, music and voice work the same way.

At a glance

At a glance
MakerByteDance
Strengths in the appNatural sounds
Who can use itPaid plans
How it’s paidCredits, from the first use
PrivacyPrivate: an open model run by NeverNo on providers chosen because they don’t keep requests. Your history stays encrypted in your browser.

Price

What the app charges, read from the same table the servers bill with. 1 credit = $0.01.

Price
Generation3 credits ($0.03)

What it is

Seed Audio 1.0 is an audio model from ByteDance, available in the Sound effects tab of NeverNo's Audio Studio. The Studio tags it natural. You type a description of a sound, a door creaking, rain on a tin roof, a crowd cheering in a distant stadium, and it returns one effect. There is no length to set and no lyrics field: the model chooses how long the sound lasts, which suits one-shot noises and short ambiences.

Its sibling in the same tab is Mirelo SFX 1.5, tagged precise, where you choose the length yourself from 2, 5 or 10 seconds. The practical split is this. Seed Audio 1.0 is for when you want a believable sound and do not mind the model deciding its duration; Mirelo is for when you need an effect to fit an exact gap in a timeline.

Be clear about the rules. Seed Audio 1.0 is a moderated model: it follows its maker's rules, and requests that ByteDance does not accept are declined. It is neither anonymized nor one of the Studio's private models; for effects served by NeverNo's own provider with no outside lab involved, use Mirelo SFX 1.5. NeverNo's own hard limits apply to every model.

Audio is paid in credits by any signed-in account, so you do not need a paid plan. The finished sound is stored in your browser, not on our servers, and you can download it from there to use in a video, a game or a podcast. Think in terms of sources and surfaces. A sound is believable when the source, the material and the room are all named: a metal spoon in a ceramic bowl, bare feet on wooden stairs, a zipper on a heavy canvas bag. Generate three variants of the same sentence and pick the one with the cleanest start and tail, since you cannot set the duration. Then trim the edges in your editor, add a short fade, and you have a file that sits naturally under dialogue or music. It is also a pleasant tool for learning sound design: compare how the same noise changes when you move it from a bathroom to a forest, from a cheap plastic toy to polished steel, and you start hearing the world as layers of material and space. Typical subjects work best: weather, household objects, vehicles, animals, crowds, doors, tools, machines and the little rustles of cloth or paper. Abstract moods, such as a sad sound, are less reliable than a named event, so describe what physically happens. Podcast hosts, indie developers and student filmmakers will find it handy for filling silences, covering a missing recording or testing how a scene feels before booking a proper session. Keep a folder of favorites sorted by theme, such as kitchen, street, workshop and weather, and your next project starts with a ready palette.

Strengths

  1. Tagged natural: effects that sound like recordings of real things.

  2. No settings to fiddle with: describe the sound and the model chooses the length.

  3. Open to any signed-in account, paid in credits, with no paid plan needed.

  4. Sits beside Mirelo SFX 1.5 in the Sound effects tab, so you can compare two takes of one idea.

Limits

  1. You cannot set the length; for exact durations of 2, 5 or 10 seconds, use Mirelo SFX 1.5.

  2. Sound effects only, with no music and no lyrics; for songs, use the Music tab.

  3. Moderated by its maker: ByteDance declines requests it does not accept, and the model is neither anonymized nor private.

When to choose it

  • Foley for a video or a podcast

    Footsteps on gravel, a kettle clicking off, pages turning, a chair scraping: the small sounds that make an edit feel real, made from a single sentence.

  • Ambience for a game or a story

    A forest at night, waves under a pier, a busy canteen. Generate a few variants, pick the most convincing, and loop it in your engine or your editor.

  • Placeholder sounds for prototypes

    Button clicks, doors, engines and alarms for an app or game prototype, ready in seconds, replaced later by final recordings if you wish.

Prompt tips

  1. Say what makes the sound and where it happens: heavy wooden door creaking open in an empty stone hall.

  2. Describe one sound at a time. If you want a layered scene, generate each layer separately and mix them in your editor.

  3. Add texture words such as distant, muffled, close or echoing; the model follows them.

Frequently asked questions

Can I choose the length of the sound?

No. Seed Audio 1.0 chooses it. Mirelo SFX 1.5 offers 2, 5 and 10 seconds.

Does it make music?

No. It makes sound effects. For music, use the Music tab of the Audio Studio.

Is Seed Audio 1.0 private?

No. It is a moderated model that follows ByteDance's rules. For effects with no outside lab involved, use Mirelo SFX 1.5. Your sounds are stored in your browser.

Do I need a paid plan?

No. Audio is paid in credits by any signed-in account.

Seed Audio 1.0 or Mirelo SFX 1.5?

Seed Audio 1.0 is tagged natural and picks its own length. Mirelo SFX 1.5 is tagged precise, with 2, 5 or 10 seconds you choose, and is private.

Try Seed Audio 1.0 in the Studio

Open the Studio