Krea's lip-sync models, read beside three published ways a mouth is driven
Krea's pricing page lists lip-sync models beside image, video and 3D work, metered together in one unit balance: 100 units a day on the entry tier, 20,000 units a month on the paid plan billed yearly, and one-time packs lasting 90 days. A commercial licence is listed on the plans; no mechanism for the lip-sync work is described. From 8 source pages, read 2026-09-12 to 2026-10-25; no maker pays to be listed.
1What these notes already record
Entries that describe what drives a mouth, at four different depths, and the two lines this register keeps about changing a performance after the clip exists.
- Audio sourceNative speech with lip-sync tied to the speaker who is on screenspeaker-aware
- Audio sourceThe card describes audio-driven cinematic video generation from an audio input with a reference image and an optional text prompthanded to the model rather than made by it
- Lip-syncLip sync and facial expressions are generated from the spoken contentdriven by the script
- A framing condition attached to lip-syncLip sync performs best when the avatar is framed closer in the scene rather than positioned far from the camera
- Lip-syncLip-sync is documentedstated without a mechanism
- Voice sourceThe speaker list returns the built-in voices and the account's custom voices, and can be filtered to either kindcatalogue or supplied sample
- Audio sourceThe accepted pairs are video with audio, video with text, image with audio and image with textsupplied or read from text
- Changing a lineLip sync keeps the picture frame for frame, voices every line again and moves the mouths to match, dropping the original ambience, at 1 credit per second on clips of any lengthpicture kept
- Lip-syncThe vendor states there is no separate dubbing step and no lip-sync patch afterwardsstated as unnecessary rather than as a feature
2What the page lists, and what it does not describe
Krea's pricing page names lip-sync models in the same breath as image, video and 3D models, and the work is metered in units that spend across all of it: a free tier granting 100 units a day, and a Pro plan carrying 20,000 units a month on the yearly billing, with packs bought once that keep for 90 days. Real-time generation, node workflows and an agent round out the description, and a commercial licence is listed on the plans.
What the page does not carry is a mechanism. A lip-sync model is named; what it takes in — audio, video, an image to drive, or the line itself — is not stated, and neither is what happens to the mouth when the input changes. The register's lip-sync column exists because documented and explained are different states, and Krea's page sits in the first of them.
3The register's four depths on the same question
Across the entries these notes hold, a mouth is decided in one of three places, and the entries admit to a fourth state where nothing is decided at all. sync-3 takes audio in and matches lip movement to the audio it is given, on footage or on a still. Synthesia generates its mouth from the spoken content, which makes lip sync and facial expression outputs of the script. MiniMax ties native speech and lip-sync to the speaker who is on screen, a statement about who is talking rather than about how the mouth is built. Kling documents lip-sync without naming a mechanism, and that is where Krea's sentence sits too.
The depths matter because they schedule differently. An audio-driven entry fixes the order of work: the take exists before the picture, and the picture accommodates it. A script-driven entry lets the line be written first and delivered last. A capability-only entry tells a production that a feature exists and leaves the workflow to be discovered, and discovery is the expensive way to learn it. PixVerse adds the other half of the same choice from the voice side, with text to speech that runs on built-in voices or on custom voices built from user-provided samples.
4The question the threads keep asking
The question printed at the top of this page is the one the collection counted under its ethics and deepfakes theme: how to tell whether a video is AI-generated, across 107 near-identical threads. Vendor documentation does not answer it, and this register does not pretend that it can. The register's own line on the subject is narrower and checkable — what a page says drives a mouth, and how much of that the reader is told.
One published statement comes closer to the question than the feature lists do: Synthesia states that lip sync performs best when the avatar is framed closer in the scene rather than positioned far from the camera. That is a page admitting its mechanism has a preferred distance, which is more useful to a director than a claim about realism — and it is the kind of condition a reader can check against their own footage.
5When the mouth is fixed after the line
A clip that already exists can have its mouth changed later. This register keeps two lines about that in the pipeline it reads: a lip-sync redub holds the picture frame for frame, voices every line again and moves the mouths to match, dropping the original ambience, while the same vendor's guides state that in the normal path there is no separate dubbing step and no lip-sync patch afterwards at all. sync-3 is documented from the other direction, taking video with audio, video with text, image with audio and image with text.
Put the three together — built with the line, driven by supplied audio, or patched onto a finished cut — and the mouth turns out to be a decision a production can make at three different moments. A page that names lip-sync models without saying which of the three they do leaves that decision open, and the cost of guessing wrong is a re-render rather than an edit.
6What each of them prints about a spoken line
| Maker | What its own page states about its models | What the plans state about use | Page read |
|---|---|---|---|
| Krea | Image, video, 3D and lip-sync models with real-time generation, Nodes workflows and the Krea Agent | Commercial licence listed on its plans | 2026-10-13 |
| SceneMixer | “Lip sync is the default: the picture stays frame for frame, every line is voiced again and the mouths are moved to match; the original ambience is dropped.” | Not stated on its own pages | 2026-10-13 |
Inclusion rule. Rows are the makers this page is about, each read on its own public page, plus one pipeline whose spoken controls this register records line by line. A maker's cell carries the wording of the page read; the pipeline's cell is one sentence quoted from its published guides. A point a page leaves open is recorded as left open rather than filled in. Order. Alphabetical by maker.
7What the collected threads counted
The line below is the question this page is named after: the collection counted it across 107 near-identical threads, under the theme it files as ethics and deepfakes.
| Question | Threads with the same question | Theme |
|---|---|---|
| How can you tell if a video is AI-generated? | 107 | ethics-deepfake |
Inclusion rule. Questions the collection counted, quoted the way they were written. Order. Descending by the thread count.
8When the other tool would serve better
- The mouth follows a supplied audio take — sync-3 matches lip movement to the audio supplied on the call, so the performance comes from the recording and the picture is fitted to it.
- The line has to be patched onto footage that exists — A lip-sync redub keeps the picture and moves the mouths to a new line, which is the route for a clip that cannot be shot again.
- The tell has to be a documented condition — Synthesia publishes a framing condition for lip sync instead of a claim about realism, which is the kind of statement a director can plan around.
Searches like "Can you tell this AI avatar is generated?" land here. Read beside it: MagicLight voice cloning and Kling motion control.