A voice from a description
Say who should be speaking (age, tone, pace, accent) and MiniMax Speech makes a matching voice, with no recording needed.
MiniMax Speech 2.8 AdAnt · by MiniMax
MiniMax Speech 2.8 is MiniMax's text-to-speech model, and you can try it here: describe a voice, paste a script, and get a natural voiceover in seconds, in HD or Turbo.

We ran this script once on each model, with no retries and no edits: a warm, friendly woman in her late twenties, casual like a creator talking to a friend: described on MiniMax, and the preset voices Sarah on Fish Audio and Jessica on ElevenLabs. Whether it sounds like a person talking rather than reading, and where it pauses.
Okay, real talk. I've tried every serum on my feed, and this is the only one I've ever finished twice. Two drops, every morning. My skin just looks awake now.
| Model | Made by | Modes | Longest text | Inputs | Test run | Test cost | Best for |
|---|---|---|---|---|---|---|---|
| MiniMax Speech 2.8 | MiniMax | Text to speech | — | Script, voice description or a saved voice | 7 s | 0.08 credits | ad voiceovers, UGC and narration |
| Fish Audio Drama 3 | Fish Audio | Text to speech, multi-speaker dialogue | — | Script with bracketed directions | 3 s | 0.01 credits | short dramas, audiobooks and emotional ad reads |
| Fish Audio S2 Pro | Fish Audio | Text to speech, multi-speaker dialogue | — | Script with bracketed directions | 2 s | 0.02 credits | even, natural reads and narration |
| Eleven v4 | ElevenLabs | Text to speech | 10,000 characters (8,000 on AdAnt) | Script with audio tags | 4 s | 0.07 credits | expressive ad reads, characters and narration |
| Eleven v4 Turbo | ElevenLabs | Text to speech | — | Script with audio tags | 2 s | 0.04 credits | drafts, long scripts and fast turnaround |
Test run and cost come from the same-script run above: the seconds until the finished file came back, and the credits for its characters. Longer scripts cost more; the try-it panel shows the price before you run.
Say who should be speaking (age, tone, pace, accent) and MiniMax Speech makes a matching voice, with no recording needed.
It handles casual phrasing, short pauses and emphasis, which is what makes a UGC voiceover believable.
HD is the most realistic; Turbo is faster and cheaper for drafts and long scripts.
Our 10-second test voiceover was ready in under 7 seconds.
A [age] [man or woman] with a [warm / bright / low] voice, [casual / confident / calm], speaking at a [relaxed / quick] pace, like [who they sound like: a creator, a coach, a narrator].
MiniMax Speech 2.8 is MiniMax's text-to-speech model, in an HD version for the most realistic voice and a Turbo version for speed. It reads a script in a voice you describe or pick, with natural pauses and emphasis. AdAnt uses it for the voiceovers in its video ads.
A clear, natural read for a product demo or a UGC-style ad.
Narration for slideshows, B-roll edits and explainers.
Make the same ad's voiceover for each market you sell in.
MiniMax Speech 2.8 is MiniMax's text-to-speech model, in an HD version for the most realistic voice and a Turbo version for speed. It reads a script in a voice you describe or pick, with natural pauses and emphasis. Our 10-second test voiceover on AdAnt was ready in under 7 seconds for about 0.08 credits.
Running this model uses paid credits from an AdAnt plan. Creating an account is free. A short voiceover costs well under 1 credit.
HD sounds the most natural and is the one this page runs; Turbo is faster and cheaper for drafts and long narration.
Yes. In AdAnt Studio you can pick from the voice library or reuse a voice you made before.
Yes, the audio you make is yours to use in your ads and videos.
MiniMax offers its speech models to developers through its own API. On AdAnt you can run it without your own provider key, from the adant-cli media command or the AdAnt plugin for Claude, paying in AdAnt credits.
VideoGoogle · fast drafts and talking clips with sound
VideoByteDance Seed · longer story clips and complex scenes with many references
VideoByteDance Seed · UGC ads and product clips with references
VideoByteDance Seed · testing many hooks and ideas cheaply
VideoTopview · longer single takes and calm, natural clips
MotionKuaishou (Kling AI) · putting a consistent character into a proven performance
ImageGoogle DeepMind · photoreal product shots and keeping a product on brand
ImageGoogle DeepMind · fast social images, thumbnails and drafts
AdAnt's agents research your market, write the script, pick the right model for each shot and make the video in one chat.
Start with AdAnt