AI Voice Generator

An AI voice generator turns written words into spoken audio. You type what should be said and describe how it should sound, such as the age, the accent, the pace and the emotion, and the model performs it. The result is an audio file you can drop into a video, a game, a podcast intro or a presentation.

On Mage the voice model is Seed Audio. It makes voices, dialogue, music and sound effects from one prompt, and returns an MP3 of exactly the length you choose: 5, 10, 30, 60 or 120 seconds. It can take one image reference, or up to three audio references to guide the sound, but not both in the same clip.

What makes it useful on Mage is what happens next. A saved character can carry a voice, and the video models that read character voices, Cherry, Raspberry, Plum, Lemon and MiniMax-H3, use it automatically, so the person you designed sounds the same in every clip. The Lipsync app can also take a voice line you made and sync a face to it.

A voice prompt has two parts: what is said and how it sounds. Put the spoken words in quotation marks so they are clearly the script, then describe the performance around them. "A warm, gravelly man in his seventies, slow and amused, close to the microphone in a quiet room, says: “Every story starts with somebody who should have stayed home.”" gives Seed Audio a speaker, a delivery and a space.

The details that change a voice the most are age, energy, pace and distance. A whisper close to the microphone, a stage voice in a large hall and a tired voice on a phone line are three different performances of the same words. Name the emotion with a verb, such as teasing, pleading or announcing, rather than an adjective alone.

• Put the exact words in quotation marks
• Give an age range and an energy level
• Name the pace: unhurried, brisk, halting
• Describe the room: booth, street, cathedral, phone
• Keep a 5- or 10-second line for testing a voice before a long read

Seed Audio returns exactly the duration you pick, so match the length to the script. A single line or a reaction fits in 5 or 10 seconds. A short voiceover or a product read fits in 30. A scene of dialogue or a narration fits in 60 or 120. A script that is too long for the clip gets rushed; one that is too short leaves silence or room tone at the end.

Plan for where the audio is going. Lipsync and Music Video take audio up to 15 seconds, so for a talking portrait, write a line that fits in 10 seconds and generate at that length.

A character on Mage can carry a voice: upload an MP3 or WAV, or generate a short sample with Seed Audio when you create the character. From then on, Cherry, Raspberry, Plum, Lemon and MiniMax-H3 use that voice automatically whenever the character speaks. The AI character generator covers characters, and the AI talking photo guide shows how to make a portrait say a line.

Use reference audio only of your own voice, or of a voice you have permission to use. Imitating a real person without their consent is not allowed on Mage.