Loading the active checkpoint information…
The base S2-Pro model can sound flatter than highly expressive TTS systems. Inline instructions may help some prompts, but V1 does not claim to solve that inherited limitation.
Use a clean, consented 10–30 second sample with one speaker and little background noise.
The transcript is required for reliable zero-shot conditioning.
Use the same seed to A/B sampling. For emotional delivery, put explicit [tag] instructions in the text.