Can you lip-sync an existing video to new audio with Sume?

No: the Sume docs we checked list no route that takes an existing video and re-syncs its mouth to a new audio track. What they do list are three other ways to end up with a speaking face. You can generate a talking video from a still image, swap a face into a clip, or generate a fresh avatar video from a script.

That is a real limit, and it is worth knowing before you plan a dubbing workflow around it. Video generation models in the router are for creating or editing footage, and the docs do not describe them as lip-syncing to a later voice-over.

What are the three documented routes?

Each starts from a different input, so pick by what you already have.

Which one is closest to dubbing?

None is true dubbing, where the original person's mouth is re-timed to a translated voice. The closest for a spokesperson workflow is to rebuild the spokesperson as an avatar and feed it the translated script. That changes the footage rather than preserving it, so it only works when the person on screen can be a generated one.

If you must keep real footage and change the words, you need a video-to-video lip-sync model outside Sume, and you should check its hardware needs first. Open models of that kind exist, with a typical cost in GPU memory, as described in the Lip Forcing comparison.

What can Sume still do for a translated clip?

For the subtitle path, see the lip-sync options overview for what each still-based route costs, and the face-swap post for what the Beta needs as input. Read the Avatar 1.0 docs for the current script, ratio and quality options before you submit.

• Burn translated subtitles onto the original video with the caption endpoint, no lip movement needed.
• Detach the original audio and split or join audio parts when you build a new track.
• Generate a new talking-head version from a still when brand rules allow an avatar.

What should you ask before choosing a route?

Ask first whether the person on screen must be the real person. If yes, an existing-footage tool is needed and you should look outside Sume for the mouth edit, then use Sume for captions and audio steps. If an avatar is acceptable, ask whether you have the photo rights and whether the script fits the 4 to 60 second window. Those two questions decide the route in most cases.

If you are not sure, run a three-second test on the cheapest route and watch it before committing a whole campaign.