I was able to spend more time playing with the AI features in KS, and while it pains me to say this, its nearly worthless to my workflow. I have no need to restyle any of my products into a steampunk version, or let any AI decide what creative direction we should go. This is already decided in our creative meetings, and is at the behest of our creative director and heās not going to accept (nor should he) any feedback from AI. Our look and feel of the images we use are carefully art directed and specific, something that AI is not capable of, and iterations are random and uncontrollable.
The back plate functions is what I was most excited about, but again, its not art directable. It is not following my guidance, and no matter the words or the prompts i use, it does not follow the camera angles, lens settings, zoom settings, etc. It gives me the wrong perspectives, wide angle vs tele, etc. I cannot tune my prompts to match the geometry. Also, the fact that it bakes the render into the background is not great, it very much limits what can be done after the render. Its too random.
What appears to have happend here, which dissapoints me a bit is that KS bought into the AI buzz and grabbed someoneās prefabbed AI and put KS clothes on it. Its just not that useful. I canāt imagine (at least not yet) that any designer worth their salary is going to use it in production. I canāt imagine in a year going back into a project that had AI elements that was made previously and need updates or changes that anything will look the same or get the same results if the AI elements need to be tweaked.
AI image generation just doesnāt have the control or predictability for production use. Sure its fun to play with, but without substantial fine controls, its just not useable. My time as a paid designer would be better used in searching adobe stock for the background plate image that matches my scene the closest and knowing intrinsically what I can and cannot fix in photoshop to make it work. AI image generation is currently fine for ethereal fantasy artwork, but as of right now its workflow is not efficient enough for production use. I donāt have 3 days (or even 4 hours) to draft a prompt that MIGHT give me the background plate I need for a scenario when I can go to a stock site and find an image in 10 min, slap it in PS, blur and desaturate it and send for approvals.
Now, I do realize that my stuff is not the prettiest, its not āsneaker on the beachā stuff, its very industrial, hard edged 80% good enough images. Most of our stuff is ābetter than what we had, which is nothingā type renders. So, AI just might not be for me, but might work great for others. Iām just a bit bummed about all the engineering hours that were spent on this, when they could have been used for other stuff that would have kept KS more on the leading edge of what its good at.
