Directing a read with text alone
You cannot tell a preset voice to 'sound excited', but you can write excitement. Short fragments read punchy. One-word sentences land. Long clauses that flow through commas read calm and assured. An ad script and a meditation script fed to the same voice come back sounding like different performers, because the engine's prosody follows sentence structure.
Lead with the voice choice, though: every preset has a natural register, and fighting it wastes takes. Audition your actual first line — not filler text — on several voices, keep the two closest, and A/B them on the full script before committing.
Timing voice over to picture
Narration runs at roughly 150 words per minute, about 2.5 words per second — so a 6-second clip carries around 15 words comfortably. Nidhogg's video models generate clips of a few seconds each, with multi-shot sequences on Seedance 2 and Kling 3, so write VO in per-shot beats: one sentence per clip, and let cuts land on periods.
Generate the voice track first when timing is tight. It is far easier to generate or trim picture against a locked read than to squeeze a re-voiced line into a locked edit — the same order most commercial editors cut in.
Replacing the voice in an existing video
If you already have footage with the wrong read — scratch audio, an outdated price, a flubbed line — generate the corrected voice over, then run the clip through Lipsync Studio. LatentSync and Sync Lipsync re-sync the on-screen speaker's mouth to the new track, so the swap does not look dubbed.
For presenter-style videos with no footage at all, InfiniTalk animates a still portrait into a talking presenter driven by your generated VO — a practical route for product explainers and course intros when there is no one to put on camera.
Alts, retakes, and series consistency
Cheap generations change ad workflows: produce three alt reads of the same script — different voices, or the same voice with restructured pacing — and test them against each other instead of guessing. When a claim or price changes, re-voice the one affected sentence and splice it in.
For a series — episode narration, a product line of demo videos — lock the engine and voice once and document the pairing. A consistent narrator across assets does more for recognition than any single perfect take.