MiniMax Speech vs Gemini Flash TTS

MiniMax Speech is 2 credits per 1,000 characters (was 3), with 5, 10, 30 or 60 second lengths. Gemini 3.1 Flash TTS is 12 credits per 1,000 characters with inline audio tags. Length versus control. Both are voice files. Neither is a talking generate.

MiniMax Speech is content type audio, category text-to-audio, audio on, durations 5s / 10s / 30s / 60s, 2 credits per 1,000 characters, catalog marked discounted. Description: MiniMax’s text-to-speech engine — clear multilingual delivery backed by one of the largest named-voice rosters in the catalog, so you can cast a specific voice rather than settle for a generic one. The picker is length. The roster is the cast.

Gemini 3.1 Flash TTS is content type audio, category text-to-audio, 12 credits per 1,000 characters, no duration chips. Description: expressive text-to-speech with 30 voices, natural-language style control, inline audio tags ([sigh], [laughing], [whispering]), multilingual synthesis, and multi-speaker dialogue. The control is inside the script. The catalog audio flag on this card is false — treat that as “not a native-audio video row,” not as “this TTS emits silence.” You still get a voice file.

If we see the mouth, you are not done. AI lipsync or a talking video row is a second bill. Text-to-speech is the mode. AI text-to-speech is the door.

Length versus control

Write which problem you actually have.

You have a 40-second explainer and you need it said in one file. MiniMax Speech. The 30s chip is too short. The 60s chip exists. Two credits per 1,000 characters at the catalog sticker. Cast a named voice from the roster. Do not split a 40-second line across two Gemini files and then stitch breaths in an editor unless you wanted that work.

You have an 8-second hook that has to laugh, then whisper the price. Gemini 3.1 Flash TTS. Inline tags and style control are the reason the card costs 12 per 1,000 characters. MiniMax’s length ladder will say the words. It will not, on this published description, parse [laughing] as performance.

You have a two-voice dialogue. Gemini’s description lists multi-speaker dialogue. MiniMax’s description lists a large named-voice roster and multilingual delivery. If the job is two people in one file with turns marked in the script, Flash TTS is the card that says that out loud. If the job is one speaker, 30 seconds, a specific named voice, MiniMax.

Do not treat 2 versus 12 as “MiniMax is cheaper TTS, always.” A 60-second MiniMax file at 2 credits per 1,000 characters is a bargain for length. An 8-second MiniMax file that cannot take the laugh you needed is a cheap wrong take. Re-rolls erase the sticker delta.

The platform-level note on tags: a separate catalog map treats Gemini, MiniMax, and Cartesia as parameter-driven in Versely’s speech tooling, while the Gemini model page still lists inline tags. Both can be true. This page follows the published Gemini description — tags are on the card — and treats MiniMax as the length-and-roster card. If a tag is spoken aloud as brackets, the take is wrong; change the row or strip the markup. Do not invent a third syntax.

Two credits per 1,000 characters, and a length ladder

MiniMax Speech is one of the few TTS rows that shows duration chips. 5, 10, 30, 60. That is a staircase of lengths, not a per-second trim. A 12-second line takes the 30s chip. A 31-second line takes the 60s chip. The bill follows the script, 2 credits per 1,000 characters of input text, not the chip, so trimming the WAV after does not refund the characters.

Fit the script to a chip when you can. A 9-second hook on the 10s step is honest. A 9-second hook on the 60s step is a long file you will cut, and a waste of the ladder’s point. The 5s chip exists for a reason: bumper, stinger, one sentence.

No resolution. This is audio. Casting from a large named-voice roster is the quality lever, not 4K. If the named voice is wrong, change the voice, not the video model.

Gemini Flash TTS has no duration enum on the published card. You do not pick 10 versus 30. You write the script and the tags. A three-minute Flash TTS dump is not “because there is no chip, it is free length.” It is a long audio generate at 12 credits per 1,000 characters that you will still have to cut. Write the line you will ship.

Multilingual is on both descriptions. That does not make either row a dubbing suite. Dubbing a video is AI dubbing plus, if we see the mouth, a lipsync pass. A 30-second MiniMax file in Spanish is a VO. It is not a replaced jaw.

Catalog MiniMax Speech is marked discounted. The sticker is 2 credits per 1,000 characters as published, discounted from 3.

Neither file is a talking generate

The expensive confusion is running TTS, laying it on a silent face, and calling the result a presenter.

Silent video plus a VO is a voiceover. Fine, if we do not see the jaw, or if you will lipsync next. Add a voiceover. Captions for silent autoplay still apply: the feed is mute. Add captions to video. Burned-in captions should match the language you synthesized.

If you needed the model to write picture and speech in one pass, you wanted native audio on a video row: Veo 3.1, Seedance 2.0, Kling O3 Pro text-to-video. MiniMax Speech will not emit 1080p. Gemini Flash TTS will not emit 4K. Twelve credits of TTS is not a cheap Veo.

MiniMax also publishes video rows (Hailuo, H3). Those are not this slug. Speech is audio. H3 is silent 2K video. Do not brief a 60-second monologue into H3 because you liked MiniMax Speech.

Photoreal cloned or named-sounding voices of real people are a disclosure problem. EU Article 50 from 2 August 2026. Article 50 disclosure. A 2-credit read in a doctor’s cadence giving medical advice is the YouTube inauthentic-content risk (synthetic personas as human authorities — Mashable, 21 July 2026) when the face or the persona is fake. The sticker does not wash it.

Pick the ladder or the tags, then stop

If the script is long and dry, MiniMax Speech, 30s or 60s, 2 credits per 1,000 characters, named voice. If the script is short and acted, Gemini 3.1 Flash TTS, 12 credits per 1,000 characters, tags. If you need both length and a laugh on a 50-second line, you may still start on MiniMax and live without the laugh, or split: Flash TTS for the hook, MiniMax for the body. Two files, two stickers, one timeline. That is cheaper than five re-rolls on the wrong row.

Run the card from the catalog. Confirm audio-only. If the screen asks for a still and 32 credits, you are in Fabric. If it asks for 5s or 10s of video, you are in a silent generate. Back out.

After the WAV: picture if you need it, captions always if it ships as video, label if it is a photoreal person.

FAQ

Which one is cheaper?

MiniMax Speech is 2 credits per 1,000 characters with a 5/10/30/60s ladder of lengths. Gemini 3.1 Flash TTS is 12 credits per 1,000 characters with tags and no duration chips. Two is less than twelve. Two is not cheaper if you needed [whisper] and MiniMax spoke the brackets.

Do I pick 5, 10, 30, or 60 on Gemini Flash TTS?

No. That ladder is MiniMax Speech. Flash TTS has an empty duration list. Write the line. Do not expect a 60-second chip.

Will either model make the face talk?

No. Both are text-to-audio. Lay the file, or lipsync it. A talking clip is a lipsync or native-audio video row.