This field of study (GAN image generation) is about 10 years old.
2 Likes
If you consider image recognition (which, for the record, is programmatically and algorithmically the same thing as image generation), it’s even older than that. You can easily go back to the 90s and see exactly the same things, or at least their predecessors
1 Like
It appears concept art and quick simple illustrations such as those used by news corporations for their various stories, are most at risk by DALLE-2. Where as 3D art is not yet at risk because we can have actual animateable obejcts and goeometry nodes systems, and we can use DALLE2 objects and turn them into 3D objects using displacement, normal and ambient occlusion mapping.
I think people can use renders as input, and then use dalle-2 to edit it almost like after FX.
(It’s a new capability they are exploring)
The image generation model - DALL-E 2 by openAI has some capabilities that a 3D artist might be interested in. Here are some of my findings: 3D capabilities of DALL-E 2
Nah Its not personalizable because it’s random.
Right now it has some very strange limitations:
Strange for an AI… it can’t count as high as 5!
While it csn impersonate Quentin Blake, it doesnt know about Pat Hutchins or Korky Paul.
It also can’t create text.
ca
when tractor come we thought farmers will lost. When computer come most of us thought every jobs are in danger. But tractor help us to do agriculture in more places and now we can feed more people. When computer come it give us more new fields of works. I hope A.I will give us more opportunities. 
1 Like
Guys with stable diffusion it can even generate hardsurface design like spaceship 
There is alot more control with img2img, you guys are only using text to image. Also inpainting is neat.
midjourney --testp is a whole different animal
“Both are for the beta model. -test is general purpose. testp is photorealistic.”
Not using Dall-E 2 or Midjourney for my spaceship sketch/design, I’m using the CEB Stable Diffusion addon for Blender(which is free on gumroad):
This is the power of img2img, with a simple shape you can turn it into a specific style. It’s just a tool, you can change things you don’t like about the output, combine,etc.
4 Likes
The fact that AI in current state have problem with details and basic shapes (like generating 4 fingers hands or ugly eyes) will be fixed with time. But AI can’t produce something meaningful like real composition where objects and characters placed and posed to express some idea. Untill it can understand idea what user want to get it will be useless for industry, it can be used for fun or as tool to help creating some textures or generic backgrounds. But looking how fast this technologies are progressing soon it will be able to generate 3d objects textured and shaded. Next step would be rigging and then posing\animating. It will be bad times for 3d artists. If they will teach AI to understand idea behind promt and will be able to generate, compose and animate objects in 3d scene exactly how user wants it will be revolution. Anyone can write some kind of script and AI will generate all assets, characters, props and then user will tell it how to animate all this stuff. For some people it will be dark days, but for majority of people it will be like heaven. But whole digital art value will be nullified, if anyone can produce top level scenes and animations in short time without any efforts, then it will have no value. On other hand majoity will be able to express their ideas just to share with friend or just for fun almost for free. Who will need then modellers, animators, riggers, painters?
NVIDIA actively exploring this field and they got really stunning result right now. I saw their AI trained with human motions can learn realtime adjusting characters reactions on different situations. I was immpressed. Also they trying to make AI more controllable by user. I think in near few years we will see revolution in CG…
I was sceptical at first but now I think it will change things drastically for all who involved in any kind of digital art.
1 Like