[read first] The new pre-alpha version of AiTube is here!

#12
by jbilcke-hf - opened

Hello everyone,

The new version of AiTube is coming now, you can already play a bit with a simplified version of it!

For the moment you can use it to generate static loops, but more improvements are planned this week to make the content less static and boring.

Regarding image quality, there are some limitations today (AiTube2 doesn't use H200 or H100 but L40S machines, meaning I had to downgrade certain settings in terms of number of steps and resolution so the image rendering quality is impacted, especially for anonymous users).

But hopefully quality should be improved through the year!


Note: do you want to see an improved quality for specific style (eg. japanese animation etc)?

If you have the know-how to train a distilled version of the LTX-Video 0.9.6 model, you are welcome to submit a demand. I'm not sure exactly how it will be implemented (the best would be for you to submit a LoRA uploaded on Hugging Face, that is already working and compatible with the 0.9.6 distilled model, then I will probably create a list of LoRA models that work well with AiTube).

For the moment the 0.9.7 version cannot be used out of the box, as AiTube requires real-time rendering to work comfortably for all users.

jbilcke-hf pinned discussion
jbilcke-hf changed discussion title from The new pre-alpha version of AiTube is here! to [read first] The new pre-alpha version of AiTube is here!

Please make discord server for it

Generations are taking a while

Sign up or log in to comment