divamgupta

Thanks a lot for the detailed feedback. We are working on some models which do not use a phonemizer

Lol

Haha interesting pivot!

The thing is that the new models keep coming every day. So it’s economically not feasible to make chips for a single model

This is our goal too.

It was not

We are still training the model. We expect the quality to go up in the next release. This is just a preview release :)

Thanks for running the benchmarks. Currently the models are not optimized yet. We will optimize loading etc when we release an SDK meant for production :)

This is just an early checkpoint. We hope that the quality will improve in the future.

Currently we don't have chunking enabled yet. We will add it soon. That will remove the length limitations.

That is our vision too!

We will try to get rid of dependencies.

We are working to fix that. Thanks

Hmm, we will look into it.

Mostly model size, and input size. Some models which use attention are O(N^2)

We have released just a preview of the model. We hope to get the model much better in the future releases.

Yup it runs on the web browser. https://clowerweb.github.io/kitten-tts-web-demo/

We will release an STT model as well.

Thanks

Thanks for posting about our project in HN! I am one of the creators of KittenTTS

Here is the link to our repo: https://github.com/KittenML/KittenTTS

I have made a Stable Diffusion implementation using Tensorflow / Keras. The weights have been ported from the original pytorch code.

This code runs about 4x faster compared to the original Torch implementation on my 8GB M1 MacBook Air. The speedup is on devices with less memory. Still gotta figure out why.

I will add it to diffusion bee soon.

I just tried it as well. I also got errors. Well, for a one-click install app, you don't need python preinstalled.

Work in progress!

Didn't realize that it will become #1 on HN. Thanks everyone.

If you are on twitter and would like to share : https://twitter.com/divamgupta/status/1569014206912929796

Well, all the other offline tools seem didn't seem very intuitive to install, for someone without technical knowledge.

Diffusion Bee is the easiest way to run Stable Diffusion locally on your M1 Mac. Comes with a one-click installer. No dependencies or technical knowledge needed.

Features: - Full data privacy - nothing is sent to the cloud - Clean and easy to use UI - One click installer - No dependencies needed - Multiple image sizes - Optimized for M1/M2 Chips - Runs locally on your computer