How to Run Irodori-TTS Anime Locally: Japanese Voice Cloning and Voice Design

Irodori-TTS-v4.1-Anime is a free, MIT-licensed Japanese text-to-speech model with 0.8 billion parameters, fine-tuned from Aratako's Irodori-TTS-v4.1-Small on independently annotated anime-style speech data. Install it from GitHub with uv, then run infer.py to clone a voice from a short reference clip, or skip the clip and design a voice from a written description instead.

What the anime fine-tune changes

The base model, Irodori-TTS-v4.1-Small, is a general-purpose Japanese TTS model. Irodori-TTS-v4.1-Anime takes the same 0.8B architecture and fine-tunes it on anime-style speech data that was annotated separately from the base model's training set. The developers note that caption conditioning and emoji controls may behave differently from the base model as a result, so a caption prompt tuned against the base checkpoint is not guaranteed to sound the same on the anime fine-tune.

Hardware requirements

Neither the model card nor the base repo's README publishes a VRAM figure for this model, so treat any specific number you see elsewhere as a guess rather than a fact. Two things narrow it down: the README lists gradient checkpointing as a memory-saving option for training, and Hugging Face hosts int8, int4, and float8 quantized checkpoints of the base model for lighter inference. If a full-precision run does not fit your card, try a quantized checkpoint before assuming the model needs more VRAM than you have.

Install Irodori-TTS

The anime fine-tune uses the same install path as the base model: clone the Irodori-TTS repo from GitHub and sync dependencies with uv. Pick the extra that matches your hardware.

Clone a voice from a reference clip

The README's own examples point --hf-checkpoint at the base model, Aratako/Irodori-TTS-v4.1-Small. Point it at phasefield-audio/Irodori-TTS-v4.1-Anime instead to get the anime-tuned voice instead of the base one; the flag downloads the checkpoint from Hugging Face on first use.

Design a voice without a reference clip

Swap --ref-wav for --no-ref to generate speech without cloning anything. Add --caption with a plain-language description of the voice you want, and the model designs one that matches instead of copying a reference recording.

Run the Gradio web UI

For clicking through options instead of editing command-line flags, the repo ships a Gradio app that exposes both voice cloning and voice design in a browser.

Prefer not to install anything first? Try the anime fine-tune in the official demo Space on Hugging Face.

Can you use it commercially?

Both the code and the weights are MIT licensed, and the model card states the fine-tune carries the same ethical guidelines as its base model. The base model's card spells those guidelines out: do not use the model to clone or impersonate a real person's voice, such as a voice actor, celebrity, or public figure, without their explicit consent, and do not use it to generate deepfakes or synthetic speech meant to mislead. Commercial use is allowed under MIT, but it is conditional on staying inside those two restrictions rather than an unqualified yes.

FAQ

How much VRAM does Irodori-TTS-v4.1-Anime need?

Not published. Neither the model card nor the base repo's README states a VRAM figure. If a full-precision run does not fit your GPU, try one of the int8, int4, or float8 quantized checkpoints on Hugging Face before assuming you need more hardware.

What is the difference between this model and Irodori-TTS-v4.1-Small?

Irodori-TTS-v4.1-Small is the general-purpose base model. Irodori-TTS-v4.1-Anime is the same 0.8B model fine-tuned on independently annotated anime-style speech data, which the developers note can make caption and emoji controls behave differently than on the base checkpoint.

Can I use Irodori-TTS-v4.1-Anime commercially?

Yes under the MIT license on both code and weights, but only within the base model's ethical guidelines: no cloning or impersonating a real person's voice without consent, and no deepfakes or misleading synthetic speech.

Does Irodori-TTS-v4.1-Anime support languages other than Japanese?

No. The model is Japanese only, same as its base model.