Edit 12/20/23
Culinarypunk is getting a new style of art in 2024, moving away from AI art as a tool in the process. As Culinarypunk shifts towards a professional product, I’ll be switching out the concept art that was created using the old style to a new one. I’ve worked on my art to find a style I can accomplish without hurting my hands, doesn't require that much cleanup from the sketching process, won’t take me weeks to complete, and lets me get across the silly ideas in my mind. It’s a little more serious looking than you may be used to but Culinarypunk’s original concept was and always will be mixing the serious with the silly, and I think the new art style will reflect this. I hope you like it, let me know what you think!The New Process
Step One
Circles and Ovals
When I start a sketch, I pull up real world references and start to break them down into shapes. First, I mark out the cross on the face and where the neck connects to the head. I then fit as many mushy circles and ovals in as I can to build out the general shape of what I want.Step Three
The Base
I use the lasso tool on a new layer to get a base. I try to keep this all the brightest color, but sometimes hair works out better when you base it in the darker color and paint in the lighter ones. I also change up the edges of the base here and there, like shrinking the jaw or slightly changing the hair/clothing flow.Step Four
Deep Shadows
Grab that lasso tool again, it's time to bring some shape into this blob. I keep my outline up and start blocking in the darkest areas, beginning with the largest and working my way down to the finer details. I can fill in a blob, like around the eye area, and carve away bits of the shadow until it looks how I want.Step Five
Midtones and Finishing Touches
Finally, I come back in with a color between the two and fill in some midtone shadows. I could stop at Step Four but I think this step adds a lot of life and personality to the art. I also pop back into the dark shadows to add any details I feel are missing if needed. I use the same carving technique with the lasso, which lets me easily add or remove areas on the layer. This is how I get sharp linework on the shadows. With that, the art is complete! Sometimes it's difficult to decide when it's too much or not enough, but if I find myself searching for more areas to add stuff, I make myself stop. At some point in the future I might bring more color into the art, but for now I'm going to keep it purple. This art style takes inspiration from old school TTRPGs, concept/character artists like Scott Flanders and Thiago Lehmann, and golden age comic books.The Old Process
There has been a ton of discourse about the use of AI art. I thought I'd showcase how I use AI like Wombo/Dream and Leonardo as tools to create my art. As somebody who is physically unable to draw anymore and having to rely on digital tools, AI has been such a savior for my hands. It's a shortcut that lets me add depth and create art in ways I didn't think was possible anymore. I understand why some people may still have hangups about AI art, but I hope this articles helps others understand why I continue to use AI art for concept art.Step One
Crappy Photoshop Painting
First, I start with a quick sketch in Photoshop. I use super broad strokes to get the general shape and idea for the image I want. It's difficult for me to get fine details without having to redo them many times, and AI tends to pop that stuff in for me, so I can save my hands a quite a bit during this step.Step Two
First AI Run
Next, I run that basic version through Wombo/Dream to change it up a bit with prompts. In this case, it put orange slices under the ears which I never would have thought of. Unfortunately, the chest is a little weird. We're going to need to fix a few things, but that's no problem because this has already saved me so much time and pain.Step Three
Back to Photoshop
Time to fix things up a bit. Things aren't quite where I'd like them yet but we're getting closer. Our little pup is getting really cute! This is where I would add any new ideas and clean up anything that I don't love.Step Four
Style Run
Finally, it's a run through a model I've trained solely on my own work through Leonardo to bring it closer to the style I use for Culinarypunk. This process can go on for hours going back and forth, but I think we've hit the right vibe already. Welcome to Culinaria, Citrus Lollie Collie!Remove these ads. Join the Worldbuilders Guild











Thank you for breaking this down and providing the resources! I'm looking to make more custom art for my own world, and knowing a good process helps immensely! How much work did you have to provide Leonardo for it to work well?
It really depends on the style for Leonardo. The style I use for Culinarypunk is based on vector art so it's already a more 'basic' style for it to grasp. The first time I ran the training it was with around 30 pictures and it did alright but I could tell it was struggling a bit. I think this might be partially due to the content though xD I can draw a potato knight, but getting Leonardo to understand what I mean by potato knight is another challenge altogether ahahahah. Right now it's been trained on 82 images I've created in this semi-vector style and I think it does a great job so I haven't uploaded more to retrain it again, but I have plans to do that before the end of the year to clean up some things and add more food-related references for it. In another world, I'm using a more realistic style so it takes me much longer to create the training art and I find Leonardo has a harder time with it overall. I'm hoping to get rid of the Wombo step in the future and only using Leonardo because the quality is so much higher, but it definitely helps with idea generation and taking ideas further.
The art looks fantastic! I'm leaning towards a 1920s art-deco style to make 'advertisments' I can sprinkle across articles to add flavor. I'm hopeful the cleaner lines and fewer colors will make it easier for an AI to work with rather than realistic images. I figure I can grab a bunch of public-domain historical ads to give it a baseline, then figure out how to give it my own framework to work off of. Thank you so much for your advice!