Why Does My AI Character Have the Wrong Accent?
Usually it's one of two things. The speech model can drop a voice's accent even when the voice is labelled with it: in our test, one accented ElevenLabs voice kept its accent on some models and sounded American on others. Or your video prompt names a language, such as "all dialogue in English", and the video model re-speaks the line in a flat, neutral accent. Choose the model by listening, and leave the language out of the video prompt.
Cause 1: the accent depends on the model as well as the voice
With ElevenLabs, a voice and a speech model are two separate choices. We found the same voice can keep its accent on one model and lose it on another.
In our test, a professional cloned voice labelled Nigerian English held its accent on eleven_v3, eleven_turbo_v2, eleven_turbo_v2_5 and eleven_flash_v2_5. On eleven_v4 and on eleven_multilingual_v2, the same voice came out sounding American.
The confusing part is that the voice's own information said both models supported that accent. Raising stability all the way didn't bring it back on the model that lost it. Running that take back through speech-to-speech onto the voice didn't either. It kept the American delivery. The labels were wrong the whole way through, and only listening told us which models worked.
Cause 2: the video prompt named a language
If your video model performs the dialogue (or plays a voice take you supply), a sentence like "all dialogue is spoken in English" can backfire. We found that naming a language made the video model re-speak an accented take in plain English. Leave it out, and the model takes its cues from the character, the setting and the take itself. That's usually what you want.
If you really need a specific language, write it into that shot's prompt yourself, on purpose, rather than letting a template add it to every shot.
How to fix a wrong accent
- Render one short line on several models. Use the same voice and the same text on each model you have access to. Keep the files side by side.
- Choose by ear. Don't trust the voice's language tags to tell you which model keeps the accent.
- Lock that model for that character. Accents drift when lines in a scene come from different models.
- Remove any language sentence from your video prompts. Check templates and presets as well as the prompts you typed.
- Listen to the final render as well as the voice take. The take can be right and the video model can still re-speak it.
An example
A Lagos-set drama has a lead whose lines sound right in the voice preview and American in the finished shot. First, check the video prompt. If it says "spoken in English", delete that and re-render one shot. If the voice preview itself sounds American, render the same line on three or four speech models and pick the one where he sounds like himself. Then use that model for every line he speaks.
In ShowMaker
Every character has a fixed voice, so a line can't wander to a different one mid-scene. Besides library voices, you can design a voice from a written description, accents included, and perform lines with delivery tags. Browse options on the AI character voices page, or see how we keep dialogue in sync with the picture.
Related questions
Why does my ElevenLabs voice lose its accent?
In our tests the accent depended on the speech model. The same voice kept it on some models and lost it on others, so render a line on each and listen.
Can I trust a voice's language tags?
Not for this. The voice we tested was tagged with its accent on a model where it still came out American.
Should I tell the video model what language to speak?
Only when you need it for that shot. A language line added to every prompt made the model re-speak accented takes in plain English.
Every fix on this page is a feature, not a workaround.
ShowMaker Studio is built around these problems — reference-locked casting, location look signatures, coverage-based scenes, delivery-timed captions, and per-shot model routing. Start free with 200 credits: narration over AI stills, no card.