chippwalters
patreon
Early thoughts on AI prompting
🕑 Added 2022-06-15 07:07:19 +0000 UTCSo I've been digging deeper into this new AI prompting and what it means for concept design and visualization.
I've decided to transfer to the Disco Diffusion AI for several reasons which I'm glad to go into detail if anyone is interested. These images shown here are all from this new DD AI. A couple of them definitely have the Syd Mead influence. I also redid the aquarium and alien octopus versions in DD to see how they compared with the midjourney AI.
And, I've been asked to illustrate a new update of a book called Blood Toys by a good friend of mine. I think there are about 50 illustrations and I want to try and do them all with AI. I originally did a number of them years ago and I will use some of my existing image prompts to help generate.
For instance, I created a generator image (above-- the one with the chainlink fence) back in 2008. And, I used that image to seed some of these others that you see above. It took me quite some time to get a formula down and I'm not sure I've got the final version yet as I'm still tweaking parameters.
So, what does this mean for the future of concept development, illustration, and visualization? Well, as I said I'm just starting out but there are some things I have already come to realize. First off, the current state of the art in this type AI is not able to do graphic design very well. Nor can it do product design. I tried unsuccessfully for quite a few hours on both of those types and really could not get anything that I thought was acceptable. I also searched both the Reddit and Discord forms for any information about either of those topics, and got nothing.
The other interesting point for me is how difficult it is to create something that you are trying to target. It certainly is great for ad hoc compositions. That said, it is much more difficult when you are specifically trying to generate a certain type of image with specified content, point of view, color scheme, and composition.
And this sort of brings things full circle too the stuff we do in Blender. It wasn't too long ago where doing photo realistic renderings was considered cheating-- with regard to concept design and illustration. And certainly anybody can create a chrome ball on a checkerboard ground plane and do a path traced render and be proud of it. But, as we all know, it takes many, many hours to really get good at modeling, rendering, and animating not to mention compositing, special effects and all the other parts that go into creating amazing 3d.
And, at least for now, it takes many hours of learning and experimenting with the AI to be able to speak to it on its level with it's own language. Not to mention, everyone will have their own particular favorite prompts and ways of putting words together which can generate a signature of sort.
So, for now, I'm looking at it as just another tool in the toolbox, just like Cycles, or Photoshop or After Effects or any other software tool that we use. And, I think those that already have a good eye, and care to put in the time, can get good at AI prompting just as those who spend the time in Blender can get good at creating wonderful renders.
Lastly, I'm interested to know if any of this stuff interests you guys, and if so let me know and I'll continue blabbing about it and perhaps make a few videos. I can certainly create a starter tute for how to get up and running with the free version of DD-- and start creating your own amazing images. If that's something people want to learn about. Just let me know in the comments.
Comments
Chipp Walters
No, I haven't yet installed it locally for a variety of reasons. I'll go into it my next post.
Joe Maron
Are you running locally? I managed to clone the Disco Diffusion repo, and got it working on my PC. In case anyone's interested, I followed the instructions here (sans the Anaconda part, as I prefer to just install Python): https://gist.github.com/MSFTserver/a05f637f32302918dd893318a4d9f62b
Jessica S
Very cool! Haven't heard of Disco Diffusion before this, but it looks like fun. Would love to see your setup and workflow in DD - as much detail as you want. :)
Samuel
I have done some John Berkey explorations and can tell that the AI recognized it's style. Syd Mead a bit more IMO.
Samuel
LOL. He is not saying that... he is saying that as now we dont have a tutorial video of DD by chipp you have a lot to choose in Youtube and nope, not missinformation or bad content, one example Quick-Eyed Sky which is creating amazing art and tutorials to name a good one. Peace! And as Chipp already knows I also want an intro and advanced or AI videos from him, just in case.
Kevin Page
yes, please! I've messed with VQGAN+CLIP a little bit, but these look much higher fidelity. As said, great for ideation!
John Brennan
Wow! That's quite a list. Albert Bierstadt (Hudson School) should be quite interesting for making surreal, dreamlike landscapes. I noticed it doesn't include illustrator Bob Peak, but does include Frank Frazetta and Boris Vallejo, but it's not a dealbreaker by any means. I'm sure the list will grow tenfold over time.
Alejandro Tejada
Yes, I want to know more about these tools! There is a lot of hype about this technology and your reviews will show them in the correct perspective. Thanks again!
Chipp Walters
There are a list of quite a few artists the AI "understands" and Berkey is one of them. Here's the complete list: https://docs.google.com/spreadsheets/d/14xTqtuV3BuKDNhLotB_d1aFlBGnDJOY0BRXJ8-86GpA/edit#gid=0
Joseph Jolton
I agree with Godzookie - good tool for getting ideas, but too awkward for doing finished work. Especially because there’s so much fudging going on in the details.
Tom Glinskas
By all means, keep chatting about these tools. It's always fascinating to find out what's going on and to see new tools, etc. I assure you, many of us are interested.
John Brennan
Nice images. Yes, please elaborate as much as possible, it's not boring by any means. All of it seems quite daunting before even beginning. I see you feed it images as a base, but I've heard others just using descriptive language. I can also see why the creators/programmers are quite protective of their AI. The Syd Mead-esque images look interesting, as they seem to capture the essence of his work. I wonder if you fed it John Berkey if that would do as well.
Chipp Walters
Yeah, I saw where it did text really well. Hard to tell when they're going to release it to the public. DD is more complicated than midjourney and there are a lot more variables to play with.
MH
IF you would like to recommend a specific best Youtube video to watch, that would be useful. Else Im not sure how your comment helps my described issue. The internet is full of videos of poor quality and misinformation. IF you are trying to register you dont want Chip to make a video, perhaps that could be done in a more straightforward way. Either way, I trust Chipp and like the way he does things so would rather a tute from him for numerous reasons, and since he was asking if people wanted one - I was giving him a yes vote.
Niklaus
While it would be cool to have a tutorial by Chipp, in the meantime, there are several just on YouTube that will walk you through the process.
ascii Kat
I love disco ! The new one 5.3 has introduced symmetry.
Joe Maron
Also very interested in this! So far, the only one I've been able to try is wombo.art (it's open to anyone, so no waiting lists). It generates interesting stuff, but I believe the target audience is different (it actually offers to mint your creations as NFT's, so there's that). Interesting times!
Richard Osborn
Great images Chipp! I'm closely following AI art as well and for graphic design, Google's Imagen might be better? It handles text much better and this trait (skill??) may lend itself to graphic design? Only problem is that, currently, it is a closed system, only available within Google.
MH
Id like to know why you switched to DD and id like an intro tute so i can start messing around with DD. Im interested in this tool but dont have the time to read all the forums and chats so an intro video tute from you so i know how to download/install and get started experimenting would really unleash me.
