Tuesday, March 12, 2024

AI news.

I found a new high speed ai chatbot on the internet called PI that is free.  It has only hallucinated once when I asked how to turn on dark mode.  It is censored but does a great job at rewriting stable diffusion text prompts.  You can try it out a few times before they make you register to continue.  If you register it will remember your previous chat conversations:

https://pi.ai/talk

I found a stable diffusion lightning model available on the web that generates an image in less than a second, as you type. It has been down a couple days now, not sure why. They say they will be back.

https://fastsdxl.ai/

[Stable diffusion 3](https://stability.ai/news/stable-diffusion-3) is in alpha. It looks like it will be amazing when it is fully released any day now.   It has features that blow away everyone else for getting the results you want. 

There is a two minute paper on it.  

I have switched over to using SDXL Lightning models from civitai.  There are several good ones.  They can give a great image in 4 steps that are 1024x1024 in size.  They do require you to also change the cfg scale down to between 1 and 2. And it is a little picky on the sampler you use, so if you get a bad result punt down to euler and try a few different ones. 

Olivio can tell you all about it


No comments:

Post a Comment