Which models generate native synchronised audio

Native audio is listed as a feature on almost every model in this category. The picker shows something more useful: which of them will let you turn it off.

Why does the toggle matter more than the capability?

Because you can't un-generate a soundtrack you didn't want.

If you've cast a voice deliberately — a licensed performer, a synthetic voice you chose, a recording you're syncing — then a model generating its own audio is producing a bed you'll be working against rather than with.

On a model with a toggle, that's one switch. On a model without one, you're mixing under audio you didn't ask for, and you can't remove it at the source.

Which inverts the usual framing. "Does this model do native audio" is a capability question with eighteen yeses. "Can I turn it off" is a workflow question with six.

Every model generates audio. Six let you decline it. That's the number that decides your workflow.

Which six have a toggle?

Four Seedance models plus two others.

Notice that the entire Seedance family has a toggle and no Google, Wan, MiniMax, Grok, LTX or Hailuo model does. That's a vendor pattern rather than a capability one.

And notice that Kling 3.0 Omni has a toggle while Kling 3.0 Turbo doesn't. Same family, different behaviour — Turbo's own description says "audio always on," which is the product telling you before you select it.

Which twelve are always on?

Everything else, and two of them are notable.

Two things worth pulling out of that.

Only three of the twelve say so in the description. Kling 3.0 Turbo names it directly. Gemini Omni Flash, Veo 3.1, Veo 3.1 Fast, HappyHorse 1.1 and Grok Imagine 1.5 all mention sound as a feature without saying it can't be disabled. The other six say nothing about audio at all — you find out in the settings panel.

Gemini Omni Flash has no duration or resolution control either. Its whole panel is references, prompt, aspect ratio and sound — and the sound reads "always on." So it's the model with the fewest decisions to make, and audio isn't one of them.

Where this falls short. "Always on" is what the picker says. Whether the generated audio is usable, or whether it can be replaced downstream in your own edit, is a separate question this post doesn't answer — and the answer differs depending on whether a licensed performer appears.

Can you just replace it afterwards?

Sometimes, and there's a hard constraint if a real person is on screen.

For a generated character, the audio is yours to handle however your edit needs.

Where a licensed human creator appears, Hexcoded's terms permit light edits only — trimming, cropping, captions, music, end-cards. Re-voicing a face outside the platform is prohibited, and it's listed as a prohibited use rather than merely a contract breach.

So "generate with audio always on, fix it later" isn't available for a licensed performer. If the voice matters and the performer is real, you need a model where audio can be switched off, or you need Talking Actors where the voice is a casting decision.

With a licensed performer, "fix the audio later" isn't a workflow. Re-voicing outside the platform isn't permitted.

How does the audio actually get generated?

For one model, Hexcoded's own documentation names the mechanism — which is more than any competitor comparison offers.

Seedance 2.5 uses an audio-visual architecture that processes visual frames and acoustic signals within the same generation pass, where traditional video generation creates silent footage first and overlays external audio afterwards. Visual movement and sound dynamics are calculated simultaneously, so footsteps, environmental ambience, surface impacts and spoken dialogue align with visible movement without a post-synchronisation step.

That's the difference between matched and synced. A matched take has nothing to reconcile; a synced one has something you maintain.

For the other seventeen models, the picker tells you whether audio is generated and whether you can disable it. It doesn't tell you how, and neither will we without a source.

How should you decide?

Three questions, in order.

All audio behaviours read from Hexcoded's Creative Studio picker in Video → Create mode in September 2026. Behaviour changes with model versions and differs by mode. Verify in the picker before relying on this.

More on model capability, access and rights in Models.