How to Fix Lip Sync Issues With AI Talking Actors

There is no sync-sensitivity slider or viseme editor to reach for on Flovaly — Talking Actors is a fully-managed generation, not a manual lip-sync tool. When a clip's mouth movement looks off, the fix almost always lives in one of three inputs you already control: the character's reference photo, the script, or the voice pick. Fixing one of those and regenerating solves the vast majority of bad clips.

Why the usual advice doesn't apply here

Most lip-sync troubleshooting guides online talk about adjusting a sensitivity setting, calibrating visemes, or fixing frame-rate export settings. None of that exists in Flovaly's Studio, and for good reason: Talking Actors doesn't composite audio onto a video you control frame-by-frame. It sends a still reference image and a generated voice track to OmniHuman 1.5, which animates the whole performance in one pass. There is nothing to manually re-time afterwards — the only lever you have is what you feed it before you hit Run.

Start with the reference photo

The single biggest factor in lip-sync quality is the character's saved reference image. OmniHuman animates directly from that still, so whatever it can't see clearly, it can't animate accurately. A photo with the face turned at an angle, partially covered (a hand, a microphone, hair across the mouth), poorly lit, or mid-expression gives the model less to work with than a clear, front-facing, well-lit shot with a neutral or relaxed expression. If a custom actor keeps producing muddy mouth movement, swap its reference photo before touching anything else — uploading a new one takes a minute, and premade stock characters were built with this in mind, which is why they tend to be the more reliable starting point when you're troubleshooting.

Script and voice: the second-biggest lever

A technically clean animation can still look wrong if what's being said doesn't match how it's delivered. Two things to check:

• Punctuation and pacing. The model leans on commas and full stops for rhythm. A long, unbroken sentence tends to produce a rushed, run-on delivery that reads as out of sync even when the mouth is technically tracking the audio. Break long thoughts into two shorter sentences.
• Accent-to-idiom match. All eight of Flovaly's curated voices are American or British — if your script uses idiom from the “wrong” side (a British voice saying “sidewalk”, an American voice saying “pavement”), the delivery lands stilted in a way that can look like a lip-sync problem even though it's a wording one. Match the voice's accent or rewrite the line.

See how to write UGC ad scripts that actually convert for more on writing lines that read as spoken rather than typed.

The one thing you don't control: the backup pipeline

OmniHuman is Flovaly's primary lip-sync engine, but it isn't the only one running behind the scenes. If OmniHuman fails on a given job — a transient provider outage, for instance — Flovaly automatically falls back to a second, separate lip-sync pipeline on a freshly generated base video, rather than failing your job outright. That fallback engine can only accept its own fixed voice, not the one you picked, so on the rare clip where it fires, you'll hear a different voice than expected even though everything else in your request was unchanged. It's not a bug in your setup — it's the system choosing to ship a slightly degraded clip over no clip at all. Regenerating usually returns you to the primary pipeline and your chosen voice.

How to actually fix a bad clip

There's no in-place editor for a finished generation — the fix is to adjust an input and run again. From the gallery, open the clip and use Re-run, which drops its exact settings back into the Studio. Swap the reference photo, tighten the script's punctuation, or try a different voice, then generate again — the new clip sits next to the old one so you can compare directly rather than guessing whether the change helped.

If the source is footage, not a photo

Everything above applies to Talking Actors, which always starts from a still image. If you're instead trying to fix lip sync on video you already filmed, you're in the wrong mode — that's Re-voice, which replaces the audio and mouth movement on existing footage rather than generating a new performance from scratch. See Face Swap vs Talking Actors if you're not sure which mode you actually need.

FAQ