Making a Vietnamese cartoon with AI voices: what works

Making a Vietnamese cartoon with AI voices works best when you treat the voices like real casting: test them side by side, keep each line short and natural, and build your animation around the audio instead of forcing the audio to fit the animation. That's the approach behind Duma Meow, my original 2D cartoon comedy about four cats who speak Vietnamese.

Here's how the process works, what each step is for, and what you can borrow if you're making your own AI-voiced cartoon.

The setup: four cats, four voices

Duma Meow stars four original cartoon cats, each with a clear personality:

• Maelis, the deadpan one
• Corin, who is always flustered
• Seraph, the wildcard
• Luna, who is always prepared

Their voices are generated in Vietnamese with ElevenLabs v3. The provisional voices are Lily for Maelis, Liam for Corin, Jessica for Seraph and Laura for Luna. "Provisional" is the key word: these aren't final picks yet. You can meet the cast in Duma Meow, four cartoon cats and a Vietnamese comedy.

Casting like it's a real audition

The core step is testing every voice on every character. I ran a 16-turn Latin-square casting test: each of the four cats said the same line, "Ham chơi!", in each of the four voices. A Latin square arranges the turns so every voice appears with every character exactly once, which makes the comparison fair. You can hear one of those tests in the ham chơi voice test Short, and I wrote about the method in how I cast voices for four cartoon cats.

If you're casting AI voices, the lesson is simple: don't judge a voice in isolation. Hear it next to the alternatives, saying the same line, attached to the same character. And keep the test line short and expressive, so personality comes through quickly.

Short lines carry the comedy

The scripts lean on short, punchy Vietnamese lines. A few that appear in the skits:

• "Bắt đầu!" (Begin!)
• "Ối giời ơi..." (Oh my god...)
• "Ủa?" (Huh?)
• "Chờ chút." (Wait a sec.)
• "Haizz." (Sigh.)
• "Không sao đâu. Ổn mà." (It's okay. It's fine.)
• "Lúc nào cũng sẵn sàng." (Always ready.)

Short lines are easier to direct, easier to time and easier to swap if a take doesn't fit the character. They also suit vertical Shorts, where every second counts. If you're learning Vietnamese, many of these lines have their own guides on this site, starting with what ối giời ơi means.

Trim, never time-stretch

Each voice line is trimmed to fit, never time-stretched. Stretching or squeezing audio to hit an exact length can warp the voice and make it sound robotic. Trimming the silence at the start and end keeps the performance intact. If a line is too long for a beat, the better fix is to rewrite the line or adjust the animation timing rather than distort the audio.

Let loudness drive the mouths

Mouth flaps are driven by loudness: when the audio gets louder, the cat's mouth opens wider. It's a simple form of lip sync that works well for 2D cartoon characters, especially in short, comic clips. It doesn't try to match exact mouth shapes for every Vietnamese sound, but it gives each line a believable rhythm and keeps the workflow manageable.

Safety checks on every clip

Every clip passes a flashing-safety check, with zero flashes per second. Rapid flashing can be a problem for viewers with photosensitive conditions, and cartoons, especially fast comedy edits, can accidentally include strobe-like moments. Checking each clip before it goes out is a small step that protects viewers.

Formats that suit short comedy

The clips are vertical 1080×1920 Shorts. The skits draw on familiar trend formats, like the slap-off challenge (cartoon, no harm), zoomies, "balloon can't drop," the expanding brain, "this is fine," the side-eye, starter packs and silent skits. Familiar formats help viewers get the joke instantly, and the Vietnamese dialogue adds the twist. Each video also gets an SEO title written in both English and Vietnamese, so it can be found by both audiences.

What you can borrow for your own cartoon

• Cast voices side by side with the same line before committing.
• Keep dialogue short and character-specific.
• Trim audio instead of time-stretching it.
• Use simple loudness-based lip sync for 2D characters.
• Check every clip for flashing before you publish.
• Write titles in each language your audience speaks.

The full series lives on the Voiceover with jimwest405 YouTube channel. If you're starting your own AI-voiced cartoon, run a small side-by-side voice test first; it's the quickest way to hear which voice actually fits each character.