Type | |
Stats | 513 2,469 |
Reviews | (97) |
Published | Jul 5, 2024 |
Base Model | |
Training | Steps: 1,000 Epochs: 1 |
Hash | AutoV2 4BABFF89E1 |
Congrats to our contest winners! Check out the amazingly terrifying entries here!
_____________________
LoRA Overview
Based Model: SDXL 1.0
Sampler: DPM++ 2M Karras (to get the effect to look like mine)
Base Prompt 1 to try: A beautiful and caustic-infused view of a wild and iridescent underwater landscape from the ancient alien future, vivid iridescent colors that induce a sense of nostalgia, indie camera angle that captures the inspiration of the surrealist scene, despite the abstractness of this moment everything is coherent and crisp, angular brush strokes, fantastical creatures schooling together and glowing the full spectrum of colors, fantastical scenery, window to a world no human has ever witnessed
Base Prompt 2 to try: A beautiful and geometrically complex view of the inside of a bedroom from the year 2500, vivid iridescent colors that induce a sense of nostalgia, indie camera angle that captures the inspiration of the surrealist scene, despite the abstractness of this moment everything is coherent and crisp, angular brush strokes, brisk weather, fantastical scenery, window to the outside world
Negative: mutated, deformed, lowres, worst quality, low quality, jpeg artifacts, error, signature, watermark, username, blurry, censorship, low quality textures, photographic, abstract, strange anatomy, out of focus
(I wrote these prompts from scratch and they are yours, free to use and remix)
Notes on the Model
This model is derived from my own art style. It is exceptional at creating brush pen art that doesn't look AI generated and it can handle a wide variety of prompts and sizes. It can be used for character designs, landscapes, abstract art and I've even gotten it to give me color paintings with some prompt massaging and img2img.
Things I've found that look great with this model: Characters, caricatures, cartoon animals, horrifying creatures, abstract/fantasy/futuristic landscapes, futuristic cityscapes, wastelands, planet surfaces, and large dramatic battles. If you're trying to get something specific- I recommend using a controlnet or other LoRAs on top. All of the black and white work I share in the images was done without any additional LoRAs or controlnet- 100% vanilla (though I did find that if I put one of my sample images against white for img2img and set it to low strength it would sometimes have an easier time with coherence if it was a specific thing I wanted to have it make for me.)
If you're struggling to get exactly what you intended, I recommend guiding the model through LoRA strength and a low strength img to img to coax the model to give you something that has color and basic form to it- then use that again as an img2img and your further outputs should look a lot closer to what you want. This of course assumes you're starting from this model. If you already have other LoRAs you like, then this model will take your outputs in a new direction.
The model was trained on 164 of my carefully selected pieces of black and white brush pen artwork (characters, faces, text, creatures, monsters, cartoons, environments, and symbols) and I have found it's best to think of this model's strength as 'form over detail.' Just as often times you see paintings that look quite abstract up close, this model seems to favor getting the overall impression of the piece to look a certain way over the specifics. For that reason, you may have good results layering in other LoRAs to help with refinement, or keep it vanilla and enjoy a more brush stroke look (personally I just bruteforced it by generating a thousand images at a time overnight and finding my favorite ones). But I've seen some work by others who have layered LoRAs on top and it looked quite good. I think you could use this model either at the beginning or end of your process. At the beginning it will help you determine the form of your output, and at the end it will turn your work into something that could have been painted or drawn with brush pens. Please share your experiments with it!
A Sample From My Dataset
A few of my comics made it into the dataset. Here is one of them :)
Ownership (Or Lack Thereof)
I'm of the belief that all art is derivative of prior art. Personally, I was inspired by Eastman and Laird's Teenage Mutant Ninja Turtles black and white comics. They in turn were inspired by Jack Kirby and Frank Miller. And on it goes. I believe artists should begin taking ownership of their styles and creating custom LoRAs so they can build a fanbase for their style (rather than letting corporations control it). By giving away my style for free, I have already been hugely inspired as people have created things I wouldn't have.
And so...
You have full rights to this model to use as you wish so make it yours and go crazy. It makes me so happy to know I'm helping other artists achieve their visions and it makes me feel like I can finally leave a legacy behind. It's my hope that this gesture can start us moving towards bridging the divide between traditional artists (like myself) and AI artists (like myself). I want to help AI artists create art that is more traditionally sound and I want to show traditional artists that the canvas they can work with has widened considerably with so much amazing new technology. I'll continue to upload my outputs from this model regularly as I have quite a few that have never seen the light of day, and I hope I can ignite a spark in each artist who uses this model to push the boundaries of art and make things that will truly inspire others- even their most ardent critics.
Want My Art on Stuff?
If you like my work enough to put it on your wall or mouse pad, I have a Redbubble! https://www.redbubble.com/people/setzstone/explore?page=1&sortOrder=recent
It can be time consuming to put my work on Redbubble and properly tag it so if you see a piece I have here and you want me to put it on Redbubble, send me a message and I'll oblige you!
If you'd rather just buy me a coffee, I do love coffee too: https://buymeacoffee.com/setzstone/steal-art-style
ā„ Remember to leave hearts and reviews ā„
For Fun
I put some drawings I did of my dog into the dataset :) He's a squatty little Dachshund-Beagle.
________________________________________________________________