This is beautiful to look at and it's nice to see some competition.
I notice some distortion like artefacts that could be improved on. When I look at the fish video example, I notice lines move that should not move. I've seen this with other GAN generations too. It's like a "stencil" layer for a texture being used in a particular region, regardless of the object's own relative positioning. Why does this happen and why don't we see this with SORA (or do we?)? Does just making it higher resolution fix this?
Maybe off the topic. But how does Chinese AI companies still have access to high end AI after Biden’s strictest export control? I presume they can’t train such a model without access to Nvidia chips?
Now that the this geni is out of bottle, China will have quickly the alternative counter part very quickly that openai. Guess LLM works now just as well in Chinese as in English?