Index-tts2 single or two person

This is the best open-source text-to-speech project until December 2025.

Workflows include single-person and two-person TTS (Text-to-Speech).

Should require 10GB of GPU memory.

First, install this custom nodes via ComfyUI:

download the models:

V2 Model Download: Manually download the models to the specified folder under ComfyUI\models\TTS:

bigvgan_v2_22khz_80band_256x bigvgan_generator.pt

config.json

-campplus

campplus_cn_common.bin

- IndexTTS-2

│ .gitattributes

│ bpe.model

│ config.yaml

│ feat1.pt

│ feat2.pt

│ gpt.pth

│ README.md

│ s2mel.pth

│ wav2vec2bert_stats.pt

│

└─ qwen0.6bemo4-merge

added_tokens.json

chat_template.jinja

config.json

generation_config.json

merges.txt

model.safetensors

Modelfile

special_tokens_map.json

tokenizer.json

tokenizer_config.json

vocab.json

-MaskGCT

semantic_codec

model.safetensors

-w2v-bert-2.0

.gitattributes

config.json

conformer_shaw.pt model.safetensors

preprocessor_config.json

README.md

Please note that Windows users may need to install wheels, such as Triton and SageAttention.

If your ComfyUI prompts you to install them,

you can: Install Triton by running the following command:

pip install -U "triton-windows<3.6"

Install SageAttention by going to Find the wheel that matches your Torch and CUDA versions and run, for example, the following command:

pip install sageattention-2.2.0+cu130torch2.9.0andhigher.post4-cp39-abi3-win_amd64.whl

other:

# 1. First, install pynini using conda (pre-compiled version, to avoid compilation issues on Windows).

conda install -c conda-forge pynini -y

# 2.

pip install WeTextProcessing --no-deps