---
license: mit
pipeline_tag: text-to-audio
tags: [text-to-audio, sound-effects]
---

# TinyPine Sound effects (EzAudio-XL)

The sound-effects model of TinyPine Studio, mirrored so the installer does not depend on third-party links.
TinyPine did not train the models.

- **EzAudio-XL** by Jiarui Hai et al. ([OpenSound/EzAudio](https://huggingface.co/OpenSound/EzAudio),
  [code](https://github.com/haidog-yaqub/EzAudio)), MIT (`LICENSE`). Changed by TinyPine: the `model` weights of
  `ckpts/s3/ezaudio_s3_xl.pt` and the `autoencoder.` weights of `ckpts/vae/1m.pt` are stored as safetensors;
  the values are unchanged.
- **FLAN-T5 XL** text encoder by Google ([google/flan-t5-xl](https://huggingface.co/google/flan-t5-xl)),
  Apache License 2.0 (`LICENSE-flan-t5`). Changed by TinyPine: only the encoder, stored in float16.
- **AudioSeal** watermark generator and detector by Meta ([facebook/audioseal](https://huggingface.co/facebook/audioseal)),
  MIT (`audioseal/LICENSE`). TinyPine embeds it in every clip it makes.

SHA-256:
- `audioseal/detector_base.pth` 8a78e8a83584113523e161fc599fcab10fd0e94c04d2eb9d2fa1e9ec91ab69d9
- `audioseal/generator_base.pth` 7a845b5fbe9364a63a3909d8ab3fe064d13a76ae4c2e983573e08c69b7b51748
- `ezaudio_s3_xl.safetensors` 77ada0a5dd0700a8113d1d6770534e2b7f5d70f8812071d1d20f04b3c7e387af
- `text_encoder/model.safetensors` 5a043d09bab6b1e17bc9b57e4636c5a54880679b87f04ed908cc73c601653cf1
- `tokenizer/tokenizer.json` fe2ebbbbde2985be723e0ce18217853e4020c5e9d35bd07be2c27ab9d3ead57a
- `vae/model.safetensors` eb85519de60954a0aae4b297cecd0828423054a7084d59d3e1062bc4835d0134