As a Ryan Zhao podcaster Put it on Bluesky“The design process went wrong when what you need for prototype is” and if there was a space. “”
I have to go fast
When Google revealed the first version of Genie earlier this year, too published a detailed search document Describing the specific stages taken behind the scenes to form the model and how this model has generated interactive videos. They did not do the same for a research article detailing the Genie 2 process, letting us guess certain important details.
One of the most important of these details is the speed of the model. The first engineering model generated its world about one framework per second, a rate that was slower orders than what would be fairly playable in real time. For Genie 2, Google only says that “the samples of this blog article are generated by an unstoppated basic model, to show what is possible. We can play a distilled version in real time with a reduction in the quality of the outputs . “
Reading between the lines, it seems that the full version of Genie 2 works with something well below real -time interactions involved by these flashy gifs. We do not know how much “quality reduction” is necessary to obtain a diluted version of the model with real -time controls, but given the lack of examples presented by Google, we must assume that the reduction is important.
The generation of IA interactive videos in real time is not exactly a pipe dream. Earlier this year, AI model Decart and hardware manufacturer Severe published The Oasis modelshowing a video clone generated by the ai-generated by the man of Minecraft It works at 20 images per second. However, this model of 500 million parameters was formed on millions of hours of sequences of a single and relatively simple game, and has been concentrated exclusively on the limited set of environmental actions and conceptions inherent in this game.
When Oasis launched, its creators have fully admitted the model “struggles with the generalization of the domain,” showing how “realistic” starting scenes had to be reduced to simplistic Minecraft blocks To get good results. And even with these limits, it is not difficult to find images oasis degenerate in horrible nightmare fuel After just a few minutes of play.
