SKAND keeps MiniMax Speech as the default for short-drama dialogue and adds optional Seed Audio performance for individual lines. Spoken text and performance direction are stored separately, and a character can bind an authorized audio reference. Leave the rest of the dialogue in place, rerecord the line that needs attention, then listen and check its timing against the shot.
Bind dialogue to a character first
Choose a voice or bind an available audio reference in the project's character settings, then assign a speaking character to each shot line. Ordinary dialogue can retain the default channel and its character settings. Pitch, emotion, cloning, and transcription controls appear only when the active provider supports them.
MiniMax Speech is the default, with APIMart speech as a fallback. Seed Audio performance is selected per line, so it does not change the default for the entire episode or cast. Availability depends on the current service configuration.
Direct an individual line
Enable performance enhancement on the selected line and enter performance direction. For example, keep the dialogue as 'I never said I was leaving' and add 'lower the voice, pause before the last word' in the direction field. The dialogue holds what is said; the direction describes delivery and rhythm.
The two fields are stored separately, and ordinary synthesis does not read those directions as dialogue. Rerecord just this line and adjust it after listening. A generated status confirms that audio is available; you still need to review wording, voice, and timing.
Use an audio reference across channels
A MiniMax clone identifier cannot be passed directly to Seed. Seed uses an authorized reference associated with the character. Without one, the interface explains that it will not preserve the MiniMax cloned voice.
References accept MP3/WAV, up to 30 seconds and 10 MB. Choose a clear passage representative of the character and confirm your rights to use it. Even with a reference, listen to results from each channel to compare voice and delivery.
Account for pauses in the price
Seed bills output duration, including pauses and generated background sound. SKAND first reserves a duration allowance for the line, then uses measured audio duration when available. If measurement cannot be confirmed, it uses the reserved line estimate; the final charge cannot exceed the reservation. One output is limited to 120 seconds.
Keep ordinary dialogue on the default channel and compare one or two important lines before changing more. Providers use different billing units, so matching character counts does not imply matching prices.
Review the result in the episode
Individual-line generation and the episode audio workflow honor each line's selected channel. Listen in the shot, check dialogue against its action, segments, and other tracks, then decide whether to rerecord or adopt it.
Speech generation is paid work. A request that has not started may receive a busy retry message; a started call with an unconfirmed outcome is not automatically submitted again. Performance direction communicates intent, but does not guarantee exact wording, voice, or emotion in every result.
Last updated: