Free guide · 7 Nov 2026

Free AI voices and lip sync for cartoon characters

How to give a cartoon character a free voice and a moving mouth: open-source voice models, three ways to lip sync, and the licence terms to check first.

By John Shemron, graphic designer and video editor at PromptVault

You can voice and lip sync a cartoon character for free with two pieces: a voice model that runs on your own computer, such as VoxCPM2 or Kokoro, and a method that turns the audio into mouth movement. For drawn 2D characters, that method is a set of mouth shapes picked from the audio, with Rhubarb Lip Sync or Adobe Character Animator's free Starter mode. For characters built in code, the mouth can simply open with the voice's loudness. Free tiers of cloud voice services work too, but they often rule out commercial use, so check the terms before you post.

This guide covers both halves, with real numbers from a finished video, and the licence terms as they stood on 7 November 2026.

Part 1: a free voice for your character

What "free" means for an AI voice

There are three kinds of free, and they come with different strings attached:

An open-source model on your computer. No account and no per-line cost. The licence decides what you can do with the audio.
A free tier of a cloud service. Easy to start, but limited by monthly credits, and often non-commercial.
Your own voice. Free, natural and yours to use. The catch is that one person voicing a whole cast takes practice.

Free voice options compared

Option	Terms (checked 7 November 2026)	Runs on	New voice from a description	Clone a recording
VoxCPM2	Apache-2.0 code and weights, free for commercial use. The project forbids impersonation, fraud and disinformation, and asks that AI content be labelled	Your computer	Yes	Yes, from a short clip
Kokoro-82M	Apache-2.0 weights. A few voices are CC BY, which means you must credit their source	Your computer; a small model	No, you pick from its built-in voices	Not offered
ElevenLabs free plan	10,000 credits a month; non-commercial use with attribution	The cloud	Yes, with a few custom voice slots	Paid plans only

Which to choose:

For a recurring character with a voice of its own, VoxCPM2 is the strongest free option, because it can design a new voice from a written description and then reuse it for every line. Its README lists 30 languages.
For a quick narrator or a light setup, Kokoro is small and has ready-made voices in several languages.
For trying ideas before you commit, the ElevenLabs free plan is fine, but the audio can't go into a monetised video without a paid plan.

How to design a character voice from a description

With a model like VoxCPM2 you write the voice, not record it. The Claude & Pixel kit (more on it below) does it in four steps:

Describe the voice in plain words. For example: "an androgynous, mid-pitched voice, neither clearly male nor female, dry and deadpan."
Generate several samples with different seeds. VoxCPM2's README says voice design "can vary between runs", so one sample isn't enough. The kit makes three or more and prints each one's pitch. As a rough guide, about 85 to 155 Hz reads as male, about 165 to 255 Hz as female, and about 145 to 195 Hz as in between.
Pick by ear, on headphones, with a line the character will actually say.
Register the winner. The chosen sample becomes the reference that every later line copies, so the character sounds the same in episode one and episode twenty.

The kit's two voices were made this way, from descriptions. They aren't copies of real people. If you clone a voice from a recording, use only a voice you have the rights to, such as your own.

Three habits that make an AI voice sound acted

Give each line a direction. A word like "deadpan" or "panicked" in front of the line changes the delivery more than rewriting the line does. The kit puts the direction in brackets before the text it sends to VoxCPM2.
Check every line with speech recognition. Voice models sometimes drop or slur a word. The kit transcribes every line with Whisper, a free speech recognition model, and if the words don't match the script, or the clip runs longer than 1.2 seconds plus 0.6 seconds per word, it tries again with a new seed.
Even out the loudness. Lines generated one at a time come out at different volumes. Normalise the finished voiceover in one pass; the kit sets it to −16 LUFS.

Part 2: free lip sync, three ways

Method	How it works	Best for	Free tools
Loudness to mouth	The mouth opens as far as the voice is loud	Characters built in code, simple or stylised mouths	A few lines of your own code
Mouth shapes from the audio	Software listens and picks one of a small set of drawn mouths for each moment	Drawn 2D characters	Rhubarb Lip Sync, Adobe Character Animator Starter
AI lip sync or a video model	A model draws the moving mouth for you	Realistic or painted one-off shots	Free tiers of AI video and lip-sync tools

Method 1: open the mouth with the voice's loudness

This is the simplest lip sync there is, and it works better than it sounds for cartoon characters. Here is how the Claude & Pixel kit does it, with the real numbers from its example video, Friday Deploy:

Measure the loudness of the voiceover 60 times a second.
Keep only the speaking character's part, so a mouth stays shut while the other character talks.
Smooth and scale it, so a loud moment is about 1 and silence is 0.
Stretch the mouth with it on every frame. In the kit, the mouth grows to as much as 3.6 times its resting height and gets a little narrower as it opens.

The first line of Friday Deploy is Pixel saying "Claude. Ship it to production." These are the values the kit wrote for it:

Word	Starts at	Loudness peak (0 to 1)	Mouth height at the peak, vs resting
Claude.	0.00 s	1.00	3.6 ×
Ship	1.08 s	0.44	2.1 ×
it	1.12 s	0.55	2.4 ×
to	1.28 s	0.96	3.5 ×
production.	1.44 s	1.00	3.6 ×

Between "Claude." and "Ship" the value drops to 0, so the mouth closes in the pause. While Claude answers "It's Friday.", Pixel's value stays at 0 and Claude's own rises. Nobody keyframed the lips; they come straight from the audio.

Friday Deploy, 47 s. Pixel's mouth follows the loudness of the voice; Claude, who has no mouth, reacts with its eyes.

The same loudness value also nods Pixel's head slightly and lifts the brows, which makes the speech look physical. Claude has no mouth at all; its eyes change shape on each line instead, and go flat for the deadpan ones.

The limit: loudness doesn't know which sound is being made, so "oo" and "ah" look the same. That's fine for a stylised character at phone size, and wrong for a close-up of a realistic mouth.

Method 2: mouth shapes picked from the audio

Hand-drawn cartoons have long used a small set of mouth drawings, swapped to match the sounds. Two free tools do the matching for you:

Rhubarb Lip Sync listens to a recording and tells you which of six basic mouth shapes (A to F), plus three optional ones, to show at each moment. It runs on Windows, macOS and Linux, writes the result as a TSV, XML, JSON or DAT file, and works with Adobe After Effects, Moho, OpenToonz, Spine and Vegas Pro. It's MIT-licensed, and its licence says the lip sync data you make belongs to you. Its default recogniser is English only; a second, phonetic one works for other languages. You draw the mouth shapes once per character, then reuse them for every line.
Adobe Character Animator in Starter mode is free with a free Adobe account; no Creative Cloud subscription needed. You pick a sample puppet or import one, record your face and voice with a webcam or import a recorded voice (AIFF, MP3 or WAV), add poses, and export an MP4. The limits: building your own puppets from Photoshop or Illustrator files and putting several characters in one scene need Pro mode, which needs a subscription.

Method 3: let an AI model draw the mouth

Video models can now generate a character speaking, with the mouth already moving. Google's Veo prompt guide covers writing dialogue into the prompt, and Kling's API has an option to generate native audio with the video. For a single painted or realistic shot, that's often the fastest route. The trade-off: the face and the voice can change from one clip to the next, which matters for a recurring character (see how to keep the same character in AI animation). For prompts that work in these tools, see AI animation prompts: 16 tested examples.

Dedicated AI lip-sync tools take an image or a video and redraw the mouth to match an audio file. Most have a free tier, but they change often. Before you rely on one, check three things: the length limit, whether it adds a watermark, and whether the free tier allows commercial use.

Which setup should you use?

A recurring series of characters built in code: a voice model on your computer and loudness-driven mouths. Free per video, and every episode matches.
A drawn 2D character: draw six to nine mouths, then run Rhubarb Lip Sync, or use Character Animator if a sample puppet suits you.
A realistic or painted one-off shot: a video model with dialogue in the prompt, or an AI lip-sync tool on a still.

For the bigger picture of these routes, see the prompt animation guide.

The setup that does it for you

The Claude & Pixel Guide sets up Method 1 end to end. You paste one prompt into Claude Code, it interviews you about the video, and you pick the voices by ear from samples it designs with VoxCPM2 on your Mac. Then it checks every line with Whisper, times every word for captions and lip sync, and renders a finished vertical video with the same two characters in every episode. It needs a Mac, and voice design needs an Apple Silicon one. It's an independent course from PromptVault, not made by Anthropic. For the format these voices are made for, see how to make animated dialogue videos.

Before you post: licences and labels

Commercial use: a monetised channel or a client video is commercial. Check the voice's terms first; the ElevenLabs free plan, for example, is non-commercial only.
Credits: some free voices, such as Kokoro's CC BY voices, require you to credit the source.
Cloning: clone only voices you have the rights to. VoxCPM2's project forbids using it for impersonation.
Platform labels: platforms treat clearly unrealistic animation differently from realistic synthetic media. How to start a faceless animated YouTube channel covers YouTube's rules for an animated series.

Frequently asked questions

Can I make a cartoon character talk for free?

Yes. Generate the voice with an open-source model such as VoxCPM2 or Kokoro, then lip sync it with Rhubarb Lip Sync, Adobe Character Animator's free Starter mode, or, for a character built in code, by opening the mouth with the voice's loudness.

What is the best free AI voice for cartoon characters?

For a character with a voice of its own, VoxCPM2, because it designs new voices from a description and keeps them the same across lines. For ready-made voices with a light setup, Kokoro. Try both with a real line from your script; the right voice depends on the character.

Can I use free AI voices on a monetised YouTube channel?

It depends on the voice. VoxCPM2 and Kokoro are Apache-2.0 and allow commercial use, though some Kokoro voices need a credit. The ElevenLabs free plan is non-commercial, so a monetised channel needs a paid plan.

How do I lip sync a 2D drawing for free?

Draw six to nine mouth shapes for your character, run your audio through Rhubarb Lip Sync to get the timing, and swap the drawings in your animation program. After Effects, Moho, OpenToonz and Spine can read its output.

Does lip sync need to match every sound?

Not for most cartoons. Simple, stylised characters read well with a mouth that just opens and closes in time with the voice. Exact mouth shapes matter more as the character gets more realistic and the camera gets closer.

The course this guide comes from

The Claude & Pixel Guide: Animated dialogue videos for Reels, Shorts and TikTok, made in code from one prompt. 15 text lessons in three tracks.

Keep reading

How to make animated dialogue videos for Shorts and Reels: How to make two-character animated skits for Shorts and Reels: the five-beat script, two voices, sitcom staging, captions and posting, with an example script.
How to keep the same character in AI animation: Why AI characters change between shots and how to stop it: a character sheet, a fixed identity block, reference images, short shots, and when to use code.
How to start a faceless animated YouTube channel: Start a faceless animated channel with two fixed characters: the series bible, an idea bank, batch episodes, hook tests, a weekly routine and YouTube's rules.