Seed Audio: Turn One Prompt Into a Full Audio Scene
Describe a scene once and get all of its sound back. With Seed Audio, a single prompt can set who's talking, how they feel, how each line is spoken, where the action happens and which sounds matter. The model turns that one description into a finished scene you can play back.
You can try Seed Audio online in a Cabina.AI chat, alongside the other audio models on the platform. The Free plan lets you start without paying. If you stay, one subscription covers every model there, so you don't pay for each one on its own.
Make Your First Scene Free on Cabina.AI
What You Can Do With Seed Audio
Full Audio Scenes
One of the main ByteDance Seed Audio features is full-scene output. Voices, SFX and ambience come back mixed in one result, so you don't have to generate separate clips and line them up in an editor yourself.
Precise Dialogue Timing
You decide when each line of dialogue comes in, and ByteDance puts the precision at 100 ms steps. Two characters can trade lines with the pauses exactly where your script puts them.
Voices That Stay Consistent
The model is built to keep a character sounding like the same person through calm narration, urgent moments and dramatic lines. You can describe a voice in words, and you can also add an audio or an image reference to guide it.
20+ Languages
Create speech in 20+ languages, from English, Spanish, German and French to Chinese, Japanese and Korean. A character can switch languages and still keep the same timbre and acting style, so the voice stays recognizable.
Clean Voiceovers
Sometimes a project needs a clear voice and nothing else. Seed Audio text-to-speech covers that case too, letting you create narration or a scripted read with no effects or background sound underneath.
Where Seed Audio Fits Your Projects
Film and TV
Build a scene with voices, room tone and effects in one clip using the Seed Audio AI generator, ready to place under your footage.
Short Drama and Animation
Characters speak their lines in the mood you write, from playful teasing to a tense standoff in an animated short.
Podcasts and Spoken Stories
A podcast intro or a bedtime story gets real emotion in the voice, so listeners don't hear a flat read.
Ads and Live Commerce
Give product promos and live commerce clips a voice and a background that match the mood of what you're selling.
Dubbing and Re-Voicing
With Seed Audio TTS, you place each line on its cue, which keeps dubbed and re-voiced scenes in sync with the picture.
Make Your First Seed Audio Clip in Four Steps
Create a Free Account
Sign up for free, and you'll have the Seed Audio generator plus the other top AI models in one place. Open a new chat once you're in.
Choose Seed Audio
Open the model list and pick Seed Audio from the audio models. If you're already in a chat with another model, you can switch to this Seed Audio generation tool right there.
Describe the Scene
Type the scene in plain words, the way you'd explain it to a friend. Then send it like any other message.
Listen, Adjust and Share
Play the result. If a line feels off, change the prompt in the same chat and run it again. When you're happy, share the chat with coworkers or friends.
What to Put in a Seed Audio Prompt
Speaker and voice. Say who is talking, such as a tired detective or a cheerful radio host, and describe the voice you'd like.
Emotion and pace. A line can be whispered, rushed or slow and calm, so tell the model how it should feel and sound.
Setting and key sounds. Picture the place for the model, like a busy cafe or a rainy street, and name the effects that should stand out.
Timing. When should each line come in, and how does the scene unfold from start to finish? Spell it out so every voice enters where you need it.
Seed Audio and Other Audio Models in One Cabina.AI Chat
Start with the words. Draft a script with Gemini or Claude, then switch to Seed Audio in the same chat. Your conversation history stays put, so the audio model picks up the script, the characters and the tone you've already worked out.
Want to hear the same script in another voice? ElevenLabs and OpenAI text-to-speech are on the same model list as the Seed Audio AI audio generator. Run one script through each model, listen to every take and keep the one you like best.
Multi-Format mode keeps text, images, audio and video in one chat. You can sketch an idea, give it a voice-over with Seed Audio and then turn it into a video with a model such as Seedance, all without leaving the conversation.
Use Seed Audio From Claude, ChatGPT or Cursor
Prefer to stay in the tools you already use? Connect Cabina.AI MCP once and call Seed Audio right from Claude, ChatGPT, Cursor or Windsurf. MCP stands for Model Context Protocol, and it acts as a secure link between your AI app and 55+ models. Builders who'd rather write code can use the Seed Audio API as well.
Setup takes a single click or one terminal command with your personal API key. You can manage, rotate or revoke keys on your own whenever you need to. Usage from every model shows up on one bill instead of separate bills from each AI vendor, and a free tier lets you test the setup first.
Claude
ChatGPT
Cursor
Windsurf
Cabina.AI MCP
Seed Audio and 55+ models
Seed Audio 1.0 Pricing and Token Cost
Seed Audio model pricing comes down to one number. Each Seed Audio 1.0 message costs 21.60 tokens, and that rate doesn't change from plan to plan. The number of messages you get depends on how many tokens your plan includes each month.
Compare the Seed Audio subscription cost across the four monthly plans.
| Plan | Price | Tokens per Month | Seed Audio 1.0 Messages |
|---|---|---|---|
| Free | $0.00 / month | 50 Cabina Tokens | 4 |
| AI | $4.99 / month | 450 tokens ($1 ≈ 90 tokens) | 20 |
| Agent | $19.99 / month | 1,000 tokens ($1 ≈ 50 tokens) | 46 |
| Agent Pro | $49.99 / month | 2,100 tokens ($1 ≈ 42 tokens) | 97 |
Seed Audio FAQ
How Is the Seed Audio Model Different From Ordinary Text-to-Speech?
Plain text-to-speech turns written lines into a voice and stops there. Seed Audio works in two stages. A language model first plans the scene from your prompt. A diffusion-based generator then renders the final audio, so the voices and the sounds around them are made together.
Does Seed Audio Support Voice Cloning?
Seed Audio voice cloning works through references. You can shape a voice from reference audio, a written description or both, and no separate model needs training for each speaker. The sample has to be one you have permission to use.
How Long Can One Seed Audio Clip Be?
A single generation runs up to two minutes, and the maker says you can continue beyond that. Long-form and multitrack output are still listed as future work, so for a full episode, plan on joining several clips.
How Does Seed Audio Compare With ElevenLabs?
Listeners in A/B tests run by the maker preferred voices that Seed Audio built from text descriptions over those of competing models. For your own scripts, your ears make the final call, and you can pick either model from the same list.
Is There Independent Proof of the Quality Claims?
The published numbers for ByteDance Seed Audio come from the maker's own tests, not from an independent lab. They report more than 90% usable audio in most scenarios and MOS scores above 4.0 for most languages. MOS measures how natural people find the speech.
Do I Need a BytePlus Account to Use Seed Audio?
No. ByteDance offers the model through BytePlus, but on Cabina.AI you use it with the same account you use for every other model. There's no extra vendor account to set up.
Can I Pay Without a Subscription?
Yes. Pay as You Go has no monthly commitment, and you can top up $3, $10, $20, $50 or a custom amount by bank card or crypto. A $20 payment gives you 1,800 tokens, and each payment includes all features for 100 days.
Try Seed Audio on Cabina.AI
Start on the Free plan and pick Seed Audio in a new chat. Once you write a scene and press send, you'll hear your first take right there.
Start Free and Hear Your Scene on Cabina.AI
