LTX 2.3 lipsync is a 20-credit mouth
LTX 2.3 lipsync is 20 credits at 1080p. Use it when the LTX picture already exists and you only need the mouth to follow a track.
LTX 2.3 lipsync is a 20-credit mouth at 1080p. Use it when the LTX picture already exists — or when the track is the brief — and you only need the mouth to follow. It is not a 4-credit redub of a phone clip, and it is not a 40-credit talking still.
LTX 2.3 Audio to Video is audio-to-lipsync. Audio on. 1080p. 20 credits. Image not required. Aspects: the family uses auto, 16:9, 9:16 on this card. Description: high-quality, fast AI video for audio-to-video with lipsync and stylized transform capabilities. The track is the brief. A start frame is optional. Write performance around the uploaded file. Do not type the VO into the prompt and upload a different take.
The sibling generate is LTX 2.3 Text to Video Fast: text-to-video, native audio optional, 4 credits, 6–20s, 4K listed. That is how you get LTX picture. This page is how you put a specific mouth on a track after — or instead of — hoping the generate heard the line.
When you already generated LTX picture
The cheap LTX habit is: generate a 4-credit Fast clip, hate the mouth, generate again. That is a lottery. The 20-credit lipsync row is the mouth pass for a track you actually approved.
Typical sequence:
- Lock picture on LTX Fast (or another LTX generate) until the blocking is right. Ignore the dummy line.
- Make or record the real VO.
- Run this 20-credit row with the track, and a start frame if identity has to match the generate.
- Caption.
If step 1 never locked, skip this row. Twenty credits will not invent a set. It will follow a track. A bad plate plus a good WAV is a good mouth on a bad plate.
Because requires_image is false, you can also start from audio alone. That is useful when the job is “this song, this performance, 1080p” and the face is allowed to be invented. It is not useful when the face is a specific presenter you already approved. Bring the start frame when identity matters. Optional is not the same as irrelevant.
Length follows the track. There is no 6s / 10s chip on this published lipsync card. Cut the WAV to the line you will ship. A 40-second ramble is not a 20-credit bargain if the picture cannot hold 40 seconds of invented motion.
The launcher is AI lipsync. The AI video generator will also list LTX. That does not make this slug a text-to-video sample. If the confirm screen wants only a prompt, you are on Fast. If it wants audio, you are here.
Twenty credits is 1080p, not 4K and not 720p
The lipsync market on Versely is a staircase of ceilings and stickers. Stay on the stair you need.
| Row | Category | Ceiling | Sticker | Still? |
|---|---|---|---|---|
| VEED Lipsync | video-to-lipsync | none listed | 4 credits | no |
| Kling Lipsync | video-to-lipsync | 4K | 7 credits | no |
| LTX 2.3 Audio to Video | audio-to-lipsync | 1080p | 20 credits | optional |
| VEED Fabric 1.0 | image-to-lipsync | 720p | 40 credits | required |
| Kling Avatar Pro | image-to-lipsync | 4K | 58 credits | required |
Twenty credits sits in the middle because the job is audio-to-video at 1080p, not because LTX is “medium quality.” If you already have footage, 4 or 7 credits is the redub. If you have a still and you need 720p or 4K identity, Fabric or Avatar Pro. If you have a track and you want LTX to drive 1080p picture, this row.
Do not pay 20 to redub a clip that already exists. That is VEED Lipsync or Kling Lipsync. Do not pay 20 hoping for 4K. The published ceiling here is 1080p. 4K is a rate lookup on the rows that list 4K.
Stylized transform in the description is not a license to treat this as Lucy Restyle. Restyle is a 1-credit look pass on footage. This is a lipsync generate driven by audio. If the job is “make this clip look like anime,” you are on an edit-video restyle row, not a 20-credit mouth.
Native audio on this card means the driving track is in the output. It does not mean LTX wrote a bed. Native audio as a one-pass generate feature lives on Fast and on other families. Here the WAV is the performance.
The track is the brief
The prompt on an audio-to-lipsync row is blocking and look, not the monologue. If you paste the script into the prompt and upload a different take, the mouth will follow the take and the prompt will argue. Pick one source of words: the file.
Record or synthesize the line as you want it heard. TTS rows exist for that. Then this row. Two bills, two jobs. Combining them into “LTX, say this” is how you land on Fast and get a mouth that almost says it.
When the picture is already LTX, keep identity continuous: same start frame, same aspect, same wardrobe in the prompt if you must prompt. A new aspect on the lipsync pass is a new crop. 9:16 versus 16:9 is a delivery choice. Make it once.
After the mouth, you still have editor work. Silent autoplay. Captions for silent autoplay. Add captions to video. Burned-in captions should match the language in the WAV, not the language you drafted in the prompt and then replaced.
Photoreal people: EU Article 50 label at first exposure for deployers of deepfakes from 2 August 2026. A 20-credit LTX mouth on a real face is still a synthetic performance of that face if the audience would take it as authentic. The template is Article 50 disclosure.
Health and finance talking heads remain a YouTube inauthentic-content problem when a synthetic persona presents as a human authority. This slug does not wash that. Know whether the face is a person, a character, or a generated extra.
Do not use this row to lottery a new scene
Twenty credits feels like a generate you can afford to spam. It is the wrong spam.
Spam Fast at 4 credits if you are hunting blocking. Spam a TTS row if you are hunting a read. Run this row once the track is the track. Re-rolling lipsync because you have not decided the line is how 20 becomes 100 without a shippable file.
If the LTX generate already has usable native-sounding speech and the line is right, you may not need this pass. Listen. If the line is wrong, do not prompt-edit the generate. Replace the mouth.
If you needed a retake of a segment inside a clip you are keeping — change the flower, replace two seconds of action — that is LTX 2.3 Retake Video at 10 credits, edit-video, not lipsync. Different row. Same brand.
Confirm audio and 1080p on the catalog. If the sticker is 4 and the input is a prompt, you are on Fast. If the sticker is 20 and the input is a track, you are on the mouth.
FAQ
Does LTX 2.3 lipsync require a still?
No. Image is not required. Bring a start frame when the face has to be someone you already approved. Skip it when the track is allowed to invent the performer.
Why is it 20 credits if LTX Fast is 4?
Fast is a text-to-video generate, 6–20s, 4K listed, optional native audio. This row is audio-to-lipsync at 1080p. Different input, different ceiling, different sticker. Four credits will not follow a WAV you upload onto this job.
Can I redub existing footage here?
You can feed a start frame and a track. If the job is “this clip, new mouth, keep the motion,” a video-to-lipsync row (VEED Lipsync at 4 credits, Kling Lipsync at 7) is the category. This card is audio-to-video, not a dedicated video-to redub.