Creator
$19/month
Billed monthly
For solo creators posting every week
500 credits every month
- Up to 250 AI images
- Up to 50 AI videos
- Up to 1.5 hours of AI voiceover
- Roll over up to 1,000 credits
MiniMax TTS. Pick the emotion, hear the read.
Sceneform runs MiniMax Speech 2.8 HD, MiniMax's text to speech model. Below, it reads one line five ways. Only the Emotion setting changes.
The samples use MiniMax preset voices. MiniMax reads in voices made by cloning or voice design. To start in your first minute, Eleven v4 reads in ready-made library voices.
0:05
The results came in this morning. It's not what we expected. We start again on Monday.
Voice Friendly Person (MiniMax preset) · Emotion neutral · Speed 1× · Pitch 0 · Volume 1 · Intensity 0 · Timbre 0 · Language Auto
Beyond emotion.Five settings, and what each one moves.
The reads above leave all five at their defaults. Move one at a time and listen before you change the next.
Speed
From 0.5 to 2, default 1. How fast the voice reads. 0.5 is half speed and 2 is double. The words stay the same.
Pitch
From -12 to 12, default 0. Moves the voice up or down in semitones. Twelve semitones is one octave.
Intensity
From -100 to 100, default 0. The energy of the delivery. Higher reads more energetic, lower reads quieter and flatter.
Timbre
From -100 to 100, default 0. The tonal quality of the voice. 0 leaves the voice as it is.
Volume
From 0.01 to 10, default 1. How loud the file comes out. 1 is the voice's normal level.
37 languages.Chinese and Cantonese are its strength.
Set Language to sharpen pronunciation for one language. Leave it on Auto and the model follows the script. Here is the same line, same voice, in both.
结果今天早上出来了。和我们想的不一样。我们周一重新开始。
0:06
Voice Friendly Person · Language set to Chinese
結果今朝出咗嚟。同我哋諗嘅唔一樣。我哋星期一重新開始。
0:05
Voice Friendly Person · Language set to Chinese,Yue
Every language you can set
AfrikaansArabicBulgarianCatalanChinese (Mandarin)CantoneseCroatianCzechDanishDutchEnglishFinnishFrenchGermanGreekHebrewHindiHungarianIndonesianItalianJapaneseKoreanMalayNorwegianNynorskPolishPortugueseRomanianRussianSlovakSlovenianSpanishSwedishThaiTurkishUkrainianVietnamese
Long scripts.Up to 10,000 characters a request.
The sample below is a short narration: 402 characters, 36 seconds, in a second preset voice, Patient Man, with every setting at its default. A full 10,000-character request at this pace runs about 15 minutes.
Every morning at six, the bakery on the corner rolls up its shutters. The first trays come out before the street lights go off. Nobody has ever written the recipe down. The owner learned it from her father, who learned it standing on a crate beside his mother. Forty years later, the bread still sells out by ten. Ask her for the secret and she shrugs. Get up early, she says, and don't rush the dough.
0:36
MiniMax or another voice model?Pick by the job and the voice.
The Audio tool runs all four. A voice you make yourself works on every one except MiniMax. See AI voice cloning.
| Model and its job | Voices | Delivery | Languages | Script |
|---|---|---|---|---|
| MiniMax Speech 2.8 HDFine control of pitch, speed and intensity, and Chinese and Cantonese | MiniMax voices from earlier cloning or design. New accounts start with none | Emotion, pitch, speed, intensity, timbre, volume | 37 | 10,000 characters |
| Eleven v4Expressive, natural narration and the widest language list | Library voices or your own | Audio tags in the script, stability, similarity | 85 | 10,000 characters |
| Seed Audio 1.0Everyday voiceovers | Preset voices or your own | Speed, volume, pitch | No setting | 2,048 characters |
| Qwen 3Chinese and English speech | Your own voices | A written style prompt | 10 | 5,000 characters |
MiniMax Speech 2.8 HD
Fine control of pitch, speed and intensity, and Chinese and Cantonese
- Voices
- MiniMax voices from earlier cloning or design. New accounts start with none
- Delivery
- Emotion, pitch, speed, intensity, timbre, volume
- Languages
- 37
- Script
- 10,000 characters
Expressive, natural narration and the widest language list
- Voices
- Library voices or your own
- Delivery
- Audio tags in the script, stability, similarity
- Languages
- 85
- Script
- 10,000 characters
Seed Audio 1.0
Everyday voiceovers
- Voices
- Preset voices or your own
- Delivery
- Speed, volume, pitch
- Languages
- No setting
- Script
- 2,048 characters
Qwen 3
Chinese and English speech
- Voices
- Your own voices
- Delivery
- A written style prompt
- Languages
- 10
- Script
- 5,000 characters
How to use MiniMax TTS.Three steps in the Audio tool.
- 1
Open MiniMax Speech 2.8 HD
The Audio tool opens with the model picked. Choose it from the model menu any time.
- 2
Pick a voice and paste the script
The voice picker lists only voices MiniMax can speak in. If it is empty, switch to Eleven v4 and pick a library voice. Paste up to 10,000 characters.
- 3
Set the delivery and generate
Choose an emotion and move speed or pitch if you need to. Generate, listen, and download the file.
MiniMax Speech 2.8 HD costs 5 credits per 1,000 characters, about 90 seconds of speech at the pace of the bakery narration.
- Creator plan on MiniMax
- 100,000 characters
- About 2.5 hours of speech
- Pro plan on MiniMax
- 260,000 characters
- About 6.5 hours of speech
- Studio plan on MiniMax
- 540,000 characters
- About 13.5 hours of speech
- Scale plan on MiniMax
- 1,100,000 characters
- About 27 hours of speech
The voiceover hours on the cards below are a general estimate at 5 credits a minute.
Simple pricing that scales with you.
Pick the plan for how much you post. Every model on every plan. Unused credits roll over. Cancel anytime.
Pro
Most popular$49/month
Billed monthly
For creators posting daily across channels
1,300 credits every month
- Up to 650 AI images
- Up to 130 AI videos
- Up to 4 hours of AI voiceover
- Roll over up to 2,600 credits
Studio
$99/month
Billed monthly
For brands and production teams
2,700 credits every month
- Up to 1,350 AI images
- Up to 270 AI videos
- Up to 9 hours of AI voiceover
- Roll over up to 5,400 credits
Scale
$199/month
Billed monthly
For agencies running always-on content
5,500 credits every month
- Up to 2,750 AI images
- Up to 550 AI videos
- Up to 18 hours of AI voiceover
- Roll over up to 11,000 credits
Every plan includesEvery model, 90+ and countingReusable AI personasSceneform Studio editorCommercial usage rights
Want checkout, comparison tables, or a custom volume? See the full pricing page
MiniMax TTS FAQ
MiniMax Speech is MiniMax's text to speech model. Sceneform runs MiniMax Speech 2.8 HD in its Audio tool. You paste a script, pick the delivery, and get an audio file back.
Plans start at $19 a month. MiniMax Speech 2.8 HD costs 5 credits per 1,000 characters, so the Creator plan's 500 monthly credits cover about 100,000 characters of speech if you spend them only on MiniMax.
It speaks in MiniMax voices made for it by cloning or voice design. To start right away, pick Eleven v4: its library voices are ready on every account, and voices you make in Sceneform read on Eleven v4, Qwen 3 and Seed Audio 1.0.
Clone it in the voice picker: record, upload or describe a voice, and it reads on every ElevenLabs model, Qwen 3 and Seed Audio 1.0. The AI voice cloning page shows how.
Speech 2.8 HD replaced Speech-02 HD in Sceneform. It takes 10,000 characters per request instead of 5,000, and it adds the Intensity and Timbre settings. Speech-02 is no longer in the model menu.
You can set 37 languages, from English, Spanish, Hindi, Arabic, Japanese and Korean to Mandarin and Cantonese. The setting boosts pronunciation for that language. Leave it on Auto and the model follows your script.
Yes. Emotion takes happy, sad, angry, fearful, disgusted, surprised and neutral. Pitch moves up or down by up to 12 semitones, speed runs from half to double, and Intensity and Timbre change the energy and tone of the voice.
Up to 10,000 characters per request. Split a longer script into parts and generate each one with the same voice and settings.
No. Sceneform runs the model for you. You sign in to Sceneform and generate in the browser, with nothing to install.
Yes. Every paid plan includes commercial use, and exports carry no watermark.
