AI Video Generator With Sound: Which Models Make Audio

If you want an AI video generator with sound, most of the video models in Wellshot can make audio with the picture: Seedance 2.5 and the Seedance 2.0 family, Seedance 1.5 Pro, Gemini Omni Flash, both Kling 3.0 models, Kling O3, Wan 3.0, MiniMax H3 Max, Grok Imagine 1.5, PixVerse v6 and FLUX 3 Video. Some let you switch sound off, some always include it, and a few are silent.

This post lists which is which, how to prompt for sound, and when to add a separate layer from Seed Audio 1.0 or the ElevenLabs models instead, including music.

Which video models make sound

In Studio there is a sound on/off switch in the prompt panel; it only does something on the models where sound is switchable.

Switchable or always on: why it matters

Switchable (the Seedance 2.x family, Seedance 1.5 Pro, Kling 3.0, Kling O3, Wan 3.0 and Wan 3.0 Prime, PixVerse v6, FLUX 3 Video) is useful when you plan to build the soundtrack yourself. Turning sound off gives you a clean picture with nothing to mute later.

Always on (Gemini Omni Flash, Kling 3.0 Turbo, MiniMax H3 Max, Grok Imagine 1.5) means every take comes with audio. If you don't want it, mute the clip on the Editor timeline and lay your own sound underneath.

Silent (Seedance 1.0 Pro, 1.0 Pro Fast and Kling 2.5 Turbo) is fine for shots that will sit under music or narration anyway, and for picture tests.

For a full head-to-head on picture quality and control as well as sound, read the best AI video models in Wellshot.

How to prompt for sound

Treat sound as part of the shot description. A line or two at the end of the prompt is usually enough.

Slow crane up over a crowded spice market at midday, rising from the stalls
to the rooftops. Bright sun, warm dust in the air, colored awnings.
Sound: crowd chatter and vendors calling, a scale clinking close by,
fading as the camera rises; wind and distant traffic at the top.

What helps:

• Tie sounds to things on screen. "A scale clinking close by" gives the model something to sync.
• Describe how the sound changes with the camera. Rising, pulling away, moving inside.
• Say what you don't want. "No music" stops the model adding a score you'll have to strip out.
• Keep spoken lines short. If you need a precise read, use a separate voice take instead (see below).

Reference sounds with Seedance

Seedance 2.5 and 2.0 can take sounds as references, not just pictures and clips. Seedance 2.5 takes up to 10 sounds, and a sound alone works as a reference. Seedance 2.0 takes up to 3, and a sound needs a picture or clip with it. On the Canvas these inputs are named @Audio 1, @Audio 2 and so on in the prompt. The Seedance reference to video guide shows how to set that up.

When to add a separate sound layer

Seed Audio 1.0, also from ByteDance's Seed team, makes speech, dialogue, sound effects and ambience in one take up to 2 minutes, from a description. Use it alongside or instead of the video model's sound when:

• The clip is silent (Seedance 1.0 Pro, 1.0 Pro Fast or Kling 2.5 Turbo).
• You need a sound bed longer than the clip, running across several cuts.
• You need a specific spoken line or narration: see the AI voiceover generator post.
• The model's sound is close but you want more control over one layer.

Details and prompts are in the AI sound effects generator guide.

Wellshot also has a set of ElevenLabs models for single jobs:

• Eleven Music makes songs and score from a description of genre, mood, instruments and lyrics, in lengths of 10 s, 30 s, 1, 2 or 3 minutes.
• Eleven Sound Effects makes one effect from a description, 2 to 22 seconds long: a door slam, a whoosh, a riser.
• Eleven v4 Voice reads a script in one of 21 stock voices, in 70+ languages; Eleven v3 Voice adds performance cues like [whispers] or [laughs]; Eleven v3 Dialogue voices a conversation between stock voices.

Prompt the video model with "No music" and lay an Eleven Music track underneath when you want the score under your control.

Mixing it in the Editor

Every take, picture and sound, lands in the project's media. In the browser Editor:

1. Drop video takes on the video tracks. Their own sound comes with them.
2. Add audio tracks for your music, voice and extra effects.
3. Use volume and mute per clip to balance the model's sound against your layers.
4. Split and slip sound to line it up with the cut.
5. Export.

If you swap a take on a shot later, the cut updates. For music, generate a track with Eleven Music or upload your own, and lay it on an audio track.

Pick a sound-on model and try your first shot: open Seedance 2.5 in Wellshot.

FAQ

Which AI video generator makes sound?

In Wellshot, Seedance 2.5, Seedance 2.0, 2.0 Fast, 2.0 Mini and 1.5 Pro, Gemini Omni Flash, Kling 3.0, Kling 3.0 Turbo, Kling O3, Wan 3.0 and Wan 3.0 Prime, MiniMax H3 Max, Grok Imagine 1.5, PixVerse v6 and FLUX 3 Video all generate sound with the picture. Seedance 1.0 Pro, 1.0 Pro Fast and Kling 2.5 Turbo are silent.

Can I turn the sound off on AI video?

Yes on the Seedance 2.x models, Seedance 1.5 Pro, Kling 3.0, Kling O3, Wan 3.0, PixVerse v6 and FLUX 3 Video. Gemini Omni Flash, Kling 3.0 Turbo, MiniMax H3 Max and Grok Imagine 1.5 always include sound, so mute the clip in the Editor if you don't want it.

Does Gemini Omni Flash generate audio?

Yes. Every Gemini Omni Flash take comes with sound, as do Kling 3.0 Turbo, MiniMax H3 Max and Grok Imagine 1.5.

Can AI generate music for my video?

Yes. In Wellshot, Eleven Music makes songs and score from a description, up to 3 minutes, or you can upload your own track. Lay it on an audio track in the Editor; video models, Seed Audio 1.0 and the ElevenLabs voice and effects models add the rest around it.