Some AI news - October 6, 2025
- Kandinsky version 5.0: A family of diffusion models for Video & Image generation has been released. Models freely available on Huggingface. (Github). I suggest you to check the samples. If you want to test these models, you can use the python command line method, but it seems that there is a ComfyUI workflow too... I didn't test the model, but it seems to be more a competitor for Wan 2.2 5b than for Wan 2.2 14b (i usually use wan 2.2 that is better). The models have a size of 4-5 GB, making it possibly interesting for people with a medium-level gamer graphic card. I only saw Text to Videos models and I didn't test this myself. However I can see that some people report out of memory problems with some configurations, even with 24gb of VRAM. This picture gives some info about the models to use with ComfyUI.
- Elon Musk wants to build "Grokipedia" (i link to the most likely url) to compete with Wikipedia. The website will be available on October 19, 2025 according to Grok. More 1, More 2.
- Finally, SORA 2 has been recently released ! Please refer to this webpage, if you want to know more. I can't test the app, because it is only available on the Apple store. For the moment, the app seems oddly uncensored.
Here are some samples that i was able to find:
Mona Lisa and the girl with a pearl earring asmr podcast with today's guest: the scream
Dora the explorer parody
Alf and ET smugling stuffs
Sponge Bob tries to escape the police
He-man and Skelator going grocery shopping
Uncensored Michael Jackson's likeness
Some japanese girl being interviewed (translate the post if you don't speak Japanese)
Japanese girl speaking in front of a computer (another point of view)
Some very convincing South Parody (the longest video)