fish-s2listed

Generate speech locally with Fish Audio S2 Pro on ComfyUI — narration and voiceover, zero-shot voice cloning from a short reference clip, multi-speaker dialogue with per-speaker tracks, and inline emotion control. 80+ languages, 44.1 kHz. Use when asked for TTS, a voiceover, narration, a cloned voice, a character voice, dubbing, an audiobook or podcast read, or whenever Fish Audio / fish-speech / S2 comes up.
ruashots/fish-s2-skill · ★ 0 · AI & Automation · score 62

Install: claude install-skill ruashots/fish-s2-skill

# fish-s2 — Fish Audio S2 Pro speech on a local ComfyUI Core philosophy: **the reference clip and the tags are the performance.** S2 Pro is a 4B model that clones a voice from 10-30 seconds and takes free-form natural-language direction inline. Almost every bad result traces to a bad reference or an untagged wall of text, not to sampling settings. ## Licence — read this before shipping anything The model is released under the **Fish Audio Research License**: research and non-commercial use only. **Commercial use requires a separate licence from Fish Audio** (business@fish.audio). This constrains monetised video, client work, and anything ad-supported. The wrapper code here is MIT; the model weights are not. ## Quick start ```bash python3 scripts/fish_s2.py check # ComfyUI, nodes, model # the model's own voice python3 scripts/fish_s2.py speak --text "Hello. [excited] This actually works." \ --name hello --normalize # make a good reference, then clone it python3 scripts/fish_s2.py prep-ref raw_recording.mp3 --out voice.wav --seconds 20 python3 scripts/fish_s2.py clone --reference voice.wav \ --reference-text "exact transcript of voice.wav" \ --text "New words in that voice. [chuckle] Including tags." \ --name narration --normalize # two voices in one pass, plus isolated per-speaker tracks python3 scripts/fish_s2.py dialogue --reference a.wav --reference b.wav --split \ --text "[speaker_1]: Did you try it? [speaker_2]: [chuckle] No, but I have opinions