Artlist AI Voiceover: Voices, Settings and Script Tips

Artlist's AI Voiceover sits in the same AI Tools as its image and video generators and uses the same credits. It offers three ways to produce speech: Text to Speech from a script, Speech to Speech from your own recording, and Custom Voices cloned from an audio sample you have permission to use. Several voice models from different providers sit behind the catalogue, and each one handles languages, accents and emotion slightly differently. This guide covers the workflow, the settings and the script habits that make the biggest difference to how natural the result sounds.

Generating a voiceover from the catalogue

1. In the side navigation, click Voiceover, and check that Voiceover is selected at the bottom of the prompt field.
2. Open the Voice Catalog from Explore All Voices. Choose Artlist Voices or your Custom Voices.
3. Filter by model, language, category (such as commercials, tutorials, documentaries or characters), gender and age, and press Play on a thumbnail to preview.
4. Click Select, then type or paste your script into the prompt area. The limit is 5,000 characters per generation for all models.
5. Choose the model and set the options it offers, then hover over Generate to see the credit cost.

Downloads are MP3. The help centre lists the catalogue's models as Eleven Multilingual v2, Eleven v3, MiniMax 02 HD and Cartesia Sonic 2 at the time of writing, and notes that Cartesia was upgraded on 1 July, so some voices may sound different from older generations.

Settings at a glance

Getting a British or Australian accent

Accent selection only works with the Cartesia model. With Eleven Multilingual v2, Eleven v3 or MiniMax 02 HD, choose a catalogue voice that already has the accent: Artlist lists British voices such as Esteem, Wit, Professor, Precision, Grounded and Posh, and Australian voices Unfiltered and Guidance.

Writing a script the voice can read

Artlist's tips for better voiceovers are specific, and following them avoids most retakes:

• Short phrases. Group phrases in square brackets for natural rhythm: [Try the new tool.] [It's easy to use.]
• Pronunciation. Match the language setting to the text, spell correctly, and spell tricky names phonetically. Write acronyms as they are spoken.
• Emphasis. Use punctuation to shape tone; Artlist suggests two question marks for questions. Do not use quotation marks for emphasis, only for actual quotes.
• Pauses. Commas or dashes for short pauses; for longer ones insert <break time="1s" /> with a native accent, or <#1.0#> (seconds) with a non-native accent. Avoid doubled punctuation.
• Numbers and codes. Break long numbers with commas or dashes, write dates in full, and wrap a code to be read character by character in <spell> and </spell>.
• Symbols, web addresses and emails. Write them as they should sound: "hashtag", "dot com".
• Generate in sections. Shorter segments are easier to pace and to fix; join them in your editor.

Directing delivery with tags

With Eleven v3 you can direct performance with audio tags: any text in square brackets before the line it should affect, in any language, for example [Calm] Good morning! or [Fast] Good morning!. Keep tags short and test one at a time.

Custom voices

A Custom Voice clones a voice from a recording. From the catalogue, click Create Custom Voice, upload a file, name the voice, add a description if you like and click Create Voice. Artlist's requirements: MP3, WAV or M4A, up to 20 MB, between 10 seconds and 5 minutes long. Toggle background-noise removal for a cleaner clone, and record in a quiet room with consistent distance from the microphone. You must confirm you have permission to use the voice; never clone someone without their consent. Custom voices can speak all supported languages except Swedish, and on team plans a voice uploaded by one member is usable by the whole team.

Speech to Speech

Speech to Speech keeps your delivery, timing and intonation but swaps the voice. Click the Speech to Speech icon in the prompt area, upload your recording (MP3, WAV or OGG, up to 30 MB and 5 minutes), choose a voice and an effect, and generate. It is the quickest route to a performance with exactly your pacing, for example matching narration to an edit you have already cut.

Keeping a project consistent

Artlist notes that every voice sounds different depending on the model, so a project should stay on one model and one voice. Record the voice, model, emotion, speed and stability alongside the script for every take you approve. If you later need a pick-up line, matching those settings is the only way to make it sound like the same session. For matching visuals, see the Artlist AI video generator guide.

Try it

[Welcome to the workshop.] [Today we're building a bird box, step by step.] <break time="1s" /> [You'll need four pieces of pine, a saw, and about an hour.]

Generate it with a British catalogue voice at speed 1.0 and a higher stability, then again with stability lowered. Listen for the trade-off Artlist describes: more expression, less consistency between takes.