This workflow uses the MOSS-TTS-v1.5 model (OpenMOSS-Team) to generate

high-quality Persian (Farsi) speech, patched with 4-bit quantization

(via bitsandbytes) so it can run on 8GB VRAM GPUs (e.g. RTX 3060 Ti).

The unquantized model (8B params, BF16) normally requires ~16GB VRAM.

Built on:

- ComfyUI node: comfyui-moss-tts by richservo

(github.com/richservo/comfyui-moss-tts)

- Model: OpenMOSS-Team/MOSS-TTS-v1.5 (Apache 2.0 license)

My contribution: added an optional 4-bit (NF4) quantization option to

the Model Loader node, which the original implementation did not

support. This patch has also been submitted upstream as a Pull

Request to the original repository.

⚠️ Disclaimer and Terms of Use:

This workflow is shared strictly for educational, research, and

entertainment purposes. Any use of it — particularly the voice

cloning capability — must be done with the consent of the voice's

owner and in compliance with the laws of the user's country. Please

do not use this tool for impersonation, fraud, or any unethical or

illegal purpose.

The creator of this post assumes no responsibility or liability for

any misuse, unethical use, or illegal use of this workflow. Full

responsibility for how it is used rests entirely with the user.