Fish Audio Drama 3 Give You A Emotive Voice TTS Locally (Patreon)
Downloads
Content
Video : https://youtu.be/4kN2ACxJ8AY
Hey everyone.
New video is up. This one is Fish Audio Drama 3 — the open 35B text-to-speech model Fish Audio built around performed speech, not reading — running natively in ComfyUI with the brand-new ComfyUI-Drama3 node pack.
We clone two character voices from a few seconds of reference audio each, save them as reusable voice files, then write a couple's argument using plain-English emotion directions — [tight, controlled], [laughs bitterly], a whispered line inside a <whisper> tag — with both speakers trading lines in one pass. After that, the finished speech gets cut into short segments and animated with MiniMax H3, characters matched to the voices.
The honest finding from my test: the first take skipped three lines of dialogue. Cause was the "..." I put at the start of a couple of lines — a leading ellipsis gets swallowed at runtime. Remove them and every word speaks. Long arguments drift; a few lines stay stable.
The free voice-cloning TTS workflow is attached below — grab it and run it.
Patreon Supporters, new member additional post for you guys in https://www.patreon.com/aifuturetech/posts/drama-3-minimax-172472893/
See you in the next one!
— Benji
Resources:
Drama-3
https://github.com/fishaudio/fish-speech
https://huggingface.co/fishaudio/drama-3
https://sgl-project.github.io/sglang-omni/cookbook/fishaudio_drama3.html
For ComfyUI
https://github.com/PedroMarinhoDev/ComfyUI-Drama3
ComfyUI Model files