Seedance 2.5 is here. Every model is included on every plan.
PricingLog inSign up
VoiceDia

Dia

Dia is a dialogue first speech model: you write a script with speaker tags and it renders the whole exchange in one take. Reach for it when the conversation, not the voice, is the thing you are making.

Make with Dia →View pricing

Added July 2026. Included on every plan.

Why this one

What makes Dia different

  1. Built for two speakersSpeaker tags in the script are the interface, so a conversation is the default rather than a workaround.
  2. One take, real timingTurn taking, overlaps and pauses are generated together instead of edited together.
  3. No voice to castThere is no voice list: the speakers come from the script structure, which removes a decision.
  4. Low cost per thousand charactersCheap enough to iterate on a scene until the rhythm is right.
  5. Natural interruptionsConversational texture that is hard to fake by concatenating separate renders.
  6. Very fast to tryWrite four lines, hear them, rewrite. The loop is short.

Use cases

Best for

  • Podcast style scenes
  • Two hander ads
  • Character banter
  • Interview mockups
  • Game dialogue
  • Comedy beats

Technique

Prompt tips

  1. Use the speaker tagsMark each line with its speaker. The tags are the whole interface, not decoration.
  2. Write like people talkContractions, fragments and interruptions. Formal prose sounds wrong coming out of a conversation model.
  3. Keep exchanges shortFour to ten lines renders tightest. Long monologues are a different model's job.
  4. Let lines overlap in the writingTrailing off and cutting in are things it can actually perform.
  5. Iterate on rhythm, not wordingMost bad takes are a pacing problem. Change where the lines break before changing what they say.

Mechanism

How Dia works

You send a script marked up with speaker tags and it renders the entire exchange as one continuous take, generating both voices and the timing between them at once. Because there is no voice picker, the casting decision is made by the model and your control is the writing, which is why the script style matters more here than on any other voice model.

Getting started

How to use it on Dolly

  1. Pick DiaChoose it in the voice studio, or just ask the agent for it by name.
  2. Write the promptDescribe what you want, in as much detail as the brief deserves.
  3. GenerateThe exact credit cost sits on the button before you press it, and a failed run refunds automatically.

Pricing

What it costs

Every generation shows its exact credit cost on the button before you run it.

per 1,000 characters
7 credits

One run is about 1/179 of a Starter month.

Failed generations refund automatically. See plans

Capabilities

Specs

Questions

Dia FAQ

Can I choose the voices?
No. There is no voice list on this model: the speakers come from the script.
How do I mark who is speaking?
With the speaker tags shown in the studio bar's placeholder. Each line carries its speaker.
How is it priced?
Per thousand characters of script, so a short exchange costs very little and a long scene scales predictably.
Can it do a single narrator?
It can, but that is not what it is for. The narration models do that better.
Why does my scene sound flat?
Usually the writing. Shorter lines, more interruption and real contractions fix most flat takes.

More like this

Other voice models

ElevenLabs Multilingual V2ElevenLabs Multilingual V2 is the classic ElevenLabs voice model, with the largest voice library on Dolly and consistently lifelike delivery.ElevenLabs V3ElevenLabs V3 is the expressive ElevenLabs model, capable of multi voice dialogue in a single generation.Gemini 2.5 Pro TTSGemini 2.5 Pro TTS is Google's studio quality speech model, with a curated voice set and multi speaker scene support.

Ready to make with Dia?

Every model is included on every plan. The price sits on the button before you run it.

Start making →