Lip Sync or Native Audio for a Talking Shot: Which Sume Route

A talking shot has two parts you need to control: what is said, and whose mouth says it. Native audio from a video model controls neither precisely, which is why Sume's docs say video models do not lip-sync to generated TTS or to a later voice-over. There are three routes that do.

The three routes

Pick by what you already have in hand.

Which to choose

Use Fabric or H3 Max lip sync when the voice is already recorded and the face is a still: a founder photo and a 12-second clip of audio is the textbook case. Use Avatar Video when the script is text and you want one avatar for up to a minute without recording. The speed_tier field is ignored on the two still-plus-audio routes, and neither offers 2K.

The audio must be hosted by Sume, so upload it first. A voice-over longer than 14.8 seconds has to be split into beats and joined afterward.

When native lip sync is enough

Kling's page describes multi-shot generation with native audio and lip sync, so a prompt-driven line of dialogue inside a Kling 3 shot is possible. The tradeoff is that the words come from the model's reading of your prompt. For brand copy that must be word-exact, a route driven by your own audio is safer.

Cost: Sume bills each route at provider list times 1.25, reserved on submit and refunded on failure. Check the live rate in GET /v1/catalog before budgeting a batch.