So I’ve been doing lots of other stuff between now and the last posting (lab reorg, server rebuilds, new tech), but I’ll cover those in another post when I can put them all together nicely! I thought I’d make a short but nethertheless interesting post on a new AI platform for image generation.
I have used Cog Video on my home lab, which is incredible in its ability to do text to image creation, so it was great interest to read about OmniGen which takes that to the next level. I loaded this up on my P100 home lab server and whilst it took a little time (c.15 min for one image, c. 30 min for two image manipulation) it was amazing to see what it could produce.
I started with some random ideas, such as breakdancing beavers in time square in new york.
I was astounded by the results, not only in image quality, but how accurate the model had turned the text into what I had asked. This was all possible with Cog VIdeo, so what is it that got me really excited about OmniGen, well it can take pictures and do a plethora of actions. One online friend suggested this was how so many of these ‘deep fakes’ where created, which raised the ethical use of AI as a benefit or a burden to society, indeed a platform like this it would be very easy to make high quality images that where so life-like it would be difficult to tell they where generated by AI.
I was netherthe less very happy to try out another AI image platform and start reading the paper so I could learn more about it myself and be able to run it on my home lab.
If you want to ‘play’ yourself, there is a platform available here, which provides access to create your own images. I’m very excited about this, but much like any ‘tool’ its unto the user to make best use of it responsibly !