AI Voice for Animation: Acting, Lip Sync and Rules
AI voice for animation covers scratch tracks, explainers and shorts. Learn the workflow, lip sync with visemes and tools, and disclosure rules for public work.
You can use an AI voice for animation as a scratch track to test timing, as narration for explainers, and sometimes as a final character voice, when your rights and plan allow it. Generate the audio, then run it through a lip sync tool such as Rhubarb or Adobe Character Animator. Check disclosure and performer rules before you publish.
Where an AI voice for animation fits in the pipeline
The table is a writer's synthesis of the sourced facts in this post, arranged by production stage.
| Stage | AI voice role | Check before release | |---|---|---| | Script and animatic | Temporary lines to test timing | Replace them if the final performance needs it | | Explainer or short | Final narration or characters, if rights and plan allow | Commercial use is included from Starter up; check disclosure rules | | Lip sync | Export the audio, then run a lip sync tool | Check the tool's supported languages and output formats | | Public release with real performers | Consent for any real performer's replica | Check the performer agreement and disclosure rules |
How to direct an AI actor: script, style and tags
Write each line as speech, not as prose. Put the mood in a short style direction and the moments in tags. Pauses are your main timing tool, so use <short pause> for a beat and <long pause> for a bigger gap before a punchline or a cut.
Animation timing depends on what the mouth does, so plan pauses with the animation in mind. Google's Cloud documentation says a specific prompt works better than a bare tag, so describing a reaction in words, such as "react with an amused laugh", can help when a tag alone is not enough.
Multi-character scenes without a recording booth
The Dialogue studio handles scenes with two characters. Each character gets a name and a voice, and a reaction inside a line, written between pipes, such as |hmm|, gives the other character a short response. The Dialogue studio accepts up to 5,000 characters per request, and names count toward that total.
Longer scenes need chunks. Google's documentation caps output audio at about 655 seconds per request, so generate the scene in parts and join the audio in your editor. For more on scripting two voices, see multi-speaker text to speech.
Lip sync with AI audio: visemes and tools
A viseme is the visual description of a phoneme, the mouth shape that goes with a sound. Microsoft's documentation describes Azure visemes as useful for "gaming avatars and cartoon characters". The table compares three tools that drive mouth shapes from audio.
| Tool | What it gives you | Notes | |---|---|---| | Azure AI Speech visemes | Viseme IDs with audio offsets (22 IDs); blend shapes at 60 frames per second with 55 facial positions | SVG output is en-US only; locale support varies | | Adobe Character Animator | A lip sync take computed from scene audio; the Chad template has 14 mouth shapes, 3 silent and 11 audio-driven | Visemes can be trimmed, swapped or set to silence | | Rhubarb Lip Sync | Six basic mouth shapes (A to F) plus optional extended shapes G, H and X; TSV, XML, JSON and DAT output | Open-source command-line tool; accepts WAV and Ogg Vorbis audio |
Rhubarb uses the PocketSphinx recognizer for English dialogue and a phonetic recognizer for other languages. Sources: Microsoft Learn on visemes, Adobe's lip sync guide and the Rhubarb repository.
OndaLabs produces the audio. This post does not claim that it outputs mouth shapes, visemes or timing data. Download the audio and use one of these tools for lip sync.
Pacing your script: a rough yardstick
The audiobook platform ACX says that "most performers narrate about 9,300 words per hour" (ACX). Dividing by 60 gives about 155 words per minute. That is our own arithmetic, not an ACX figure, and it is a narration yardstick, not a standard for character dialogue. Use it to rough out an animatic, then time the real audio. For budgeting, OndaLabs' own estimate is about 900 characters per minute on Onda Lite.
Rules to know before you publish
YouTube requires creators to disclose realistic content that is altered or synthetic, when it "Makes a real person appear to say or do something they didn't do". Its examples of content that does not need that disclosure include "Someone riding a unicorn through a fantastical world", and the exemption depends on the content not looking realistic. For non-music AI audio or animated content, a label can appear in the expanded description. Disclosing does not limit a video's audience or its monetization eligibility (YouTube).
Union rules for performers are a separate question. The 2025 SAG-AFTRA video game agreement sets consent and disclosure terms for digital replicas. It is a games contract, not an animation contract, and we did not verify animation union terms. For voice and likeness law, see voice cloning laws.
A simple workflow from script to final cut
- Write the script in speech-like lines, and rough out the timing with the yardstick above.
- Pick a prebuilt voice for each character, or design or clone a voice for a recurring character when you have permission to do so.
- Direct each line with a short style and tags, and use pauses to set the timing.
- Generate short scenes, and listen to each line before you continue.
- Download the audio and run it through a lip sync tool.
- Replace temporary lines with final performances where the project needs them, and check disclosure rules before release.
Make animation voices with OndaLabs
The free plan gives you 2,000 credits a month, about 2 minutes of audio, which is enough to test a short scene on Onda Lite with tags. Starter, at $9 a month, includes 170,000 credits, about 189 minutes, plus Onda Flash, image generation and commercial use. You can compare the rest on the pricing page. To start, sign up and build a scene in the Dialogue studio or the text to speech studio.
Frequently asked questions
Can I use AI voices for an animated video or cartoon?
Yes, within your plan and the rules of the platform where you publish. You can use AI voices for scratch tracks, explainers and short animated pieces. For final character voices, check commercial use terms, disclosure rules and any performer agreements that apply.
How do I lip sync an animated character to AI audio?
Generate the audio, then run it through a lip sync tool. Rhubarb reads WAV or Ogg Vorbis audio and outputs mouth shapes in several formats. Adobe Character Animator can compute a lip sync take from scene audio. OndaLabs does not claim to output mouth or timing data.
What are visemes?
A viseme is the visual description of a phoneme, the mouth shape that goes with a sound. Azure AI Speech can return viseme IDs with audio timing, which animators use to drive a character's mouth. Locale support varies, and SVG output is en-US only.
Should I use AI voices as scratch tracks before hiring actors?
Scratch tracks let you test pacing and story before you book a session. Keep them clearly temporary, and plan to replace them with final performances if the project needs that. This is general advice about production planning, not a claim about how any studio works.
Do I have to disclose AI voices in animated videos on YouTube?
YouTube's disclosure rule covers realistic altered or synthetic content that makes a real person appear to say or do something they did not. Its page says content that does not look realistic does not need that disclosure, though a label can appear in the expanded description for animated content. Read the current policy before you upload.