I am pretty impressed. I uploaded some elevations, a few prompts, and here are the results. I also ran it through Magnific AI.
Now, it doesn’t look like the elevations, but in a few years it’ll probably be spot on. I was also able to change materials, and add/remove detail. At first it created a curb cut at the sidewalks and I asked CHAT GPT to remove it, and it did exactly as I asked.
It’s def a step forward. The image does not look as good as render yet, but I’m sure its a matter of time. Then you can use other ai’s to mask remove, add objects to specific areas and create animations out of it, like in runway. Just pains to see 3ds max and other software fade away into the void over time…kind of scary really.
The first thing I thought was, this took me minutes and I loose hours with MAX crashes. We pay software companies to innovate, not sleep at the wheel. I could come out of 20 year coma and MAX wouldn’t be any different to me. Seeing AI create renderings with a prompt isn’t new, but building from elevations, in seconds, is wild to see.
I’m not sure about this. Clients want visualizations of their products not “generic” products. AI don’t know how the products will look like, because they don’t exist yet.
Anyway, the rendering looks nice, but the design of the house is unrealistic.
That is in beta right now, correct?
WHat I find AI does really well is where objects meet…it has real world detail where CG often looks too clean or fake. It also does grass and dirt really well. Ive found a good combo of midjourney edit painting, and magnific to get high res results. Especially with motion blur on automotive work, we can make some huge set extensions really quickly when it used to be a huge headache to add the bleed clients want out of nothing.
yep.
adobe cc pro account includes beta app access.
what nano appears to do really well is keep details from the previous image. I’ve not seen this work as well with any other AI model.
In-painting to change specific details is really easy as well.
there are a few things that are a bit weird still, like it seems to stretch the image horizontally each time you make a request when editing a full image, but it is in beta so I guess these bugs will be ironed out pretty quickly.
I’ve tried it out on various source images, basic model screen shots, hand sketches, 2D elevations, turning full renders into concept watercolours etc. and it does a pretty good job on all of them.
not sure where that leaves those of us who have spent the last 20 or so years honing our craft tbh?
it requires basically no skill to use AI.
maybe Adobe have screwed themselves here as it’ll take a lot less people to do the job now, which means less licenses, and less revenue?
I don’t know. it’s been a pretty depressing day…
Hey S0real - was this generation based on a 3D model, or did it create a 3D view based on the elevation?
the first “3D” image was based on the flat elevation and the subsequent images based off that.
Yeah nano banana is crazy good. I don’t even place 3D dudes in my scene anymore too much work of rendering in higher resolution to prevent the mangled people from afar. I just sketch a little dude on my visual to give it a sense of scale, position and pose, then describe which type of person needs to stand/sit and which clothes they are wearing and what they are doing, and it follows the prompt perfectly, perspective, lighting, shadows, reflections, near perfection, or at least a hell of a lot better than what I can do in that short time.
the harmonize feature is pretty good for adding people, just drop in a character from your people library and click the ‘harmonize’ button, and photoshop will match the lighting, add shadows/reflections. It gives 3 lighting options to choose from.
then you can use nano banana to adjust poses. (make this guy turn to face the camera)
Thanks S0real - I was afraid you would say that… I’m considering dog grooming as a new career 
Nano Banana can still only do 1200px max res. You can obvs scale that up using topaz, or Magnific, but there are still MAJOR issues in terms of quality. We are getting away with it on some shots because the backplates have motion blur, but if they are static shots, its a different matter. When clients need a 6000px car in a 12,000px backplate, with 30% bleed, you can be talking an 18,000px photoshop comp. Good luck Nano Banana-ing your way out of that problem on a stationary product shot with no motion blur.
If all of your work is sub 4k, then yes, Id be worried, but I dont think Ive ever had a final image that was sub 4k, so AI is still way way off the pace for creating final content that can compete. Perhaps one day it will get there, but I see high res content as so niche, as developers might not even see it as something to strive for.
not sure it’s way off being able to do hires stuff. i assume most people are using 4k minimum displays, apple studio display is 5k and monitors are only ever going to get better so the demand for hires will be there.
I guess it depends on your particular CGI discipline as to how long it’ll be before your job can be done by an AI.
more niche / specialist stuff is always going to be harder to do with AI than generic stuff.
Maybe AI will still be just another tool that speeds up the work in a lot of cases, but for some aspects of CGI it’s definitely going to replace people.
I wonder what Chaos group’s take is on this stuff? I imagine it’s going to have an impact on sales in the next few years… will they lean into AI and develop those tools further? It’s already happening with stuff like the material generator in Cosmos for example.
This one blew me away - Artlist
Bye bye CGI industry.
1 input frame?
Just a still image with a prompt
