This LoRA is designed to bring the unique charm of expressive sketches and textured doodles into your generations.
This style is trained on a selection of carefully chosen sketches and digital drawings. It captures the beauty of visible linework and the texture of hand-drawn art, combined with bold, deliberate color choices.
Expect visuals that feel both spontaneous and artfully composed. The model excels at creating images with clear, confident outlines, textured shading, and distinct color palettes that give each piece a strong emotional mood. Characters and objects retain a classic, illustrative quality, as if lifted from an artist's personal sketchbook but finished with a painterly touch.
The goal of this LoRA is to replicate that authentic "doodle" feeling: a mix of rough, energetic lines and beautifully structured composition. If you're looking to create illustrations that radiate personality, charm, and a handcrafted feel, this LoRA will help you achieve it.
For those who want custom-tailored styles or specific commissions, I also offer LoRA commission work. You can reach out to me directly via Private Message or visit my DeviantArt page:
👉 https://www.deviantart.com/estylonshop
Enjoy experimenting with this model, and don’t forget to share your results—I’d love to see what you create
Description
FAQ
Comments (12)
This is WONDERFUL! Please tell us some tips for style training! Which software did you use for training? How did you caption each image in the dataset?? @Estylon
https://estylon.substack.com/p/teaching-krea-2-to-draw-like-akira
natural language for Krea 2, json prompts for Ideogram 4
@Estylon I have gifted you some buzz for replying to me! Thank you!
Did you use the main branch of AI toolkit? Or a branch of it?
Also, you simply tagged every image with "Minimal Sketch Style"?
No other description or tags? Just a trigger token?
Also, how many images did you use for the Toriyama one?
How many images for the minimalist one?
What graphics card did you train on?
How long did the total training take?
ANY advice helps!
And your article is wonderful btw
@wktraÂ
Main branch, tagged every image with caption + trigger token, 100ish image for Toriyama it's from a scan of an artbook, minimalist one should be between 30 and 50 can't remember now.
Trained on 4090 for 1 hour and half.
@Estylon I'm grateful for your reply!
One more annoying question because this is what prevents good loras from being created:
- can you please give me an example of what a caption looks like, so I know how basic or how detailed an image has to be when training styles.
@wktra "The image is a black and white digital drawing of a young person likely a teenager or young adult depicted in a minimalist sketchy style. The subject is shown in profile facing to the left with their eyes closed and a serene expression on their face. They have short tousled hair with a few strands blowing to the side suggesting a slight breeze or movement. The person is wearing a hooded sweatshirt with the hood pulled over their head adding to the casual relaxed vibe of the drawing. The lines are loose and sketchy giving the impression of a quick spontaneous sketch rather than a detailed finished piece. There are no additional objects or background elements making the subject the focal point of the image. The overall mood conveyed is one of tranquility and introspection."
@Estylon THANK YOU!!!!!
@Estylon OMG, I hope I didn't just screw myself up, but I was ONLY requesting help for Krea2. I have no interest in Ideogram4! ðŸ˜
@wktra it's ok i literally written the stuff related to Krea 2.
@Estylon I'm using your guide to train a lora. I'm curious why you chose diffusers and not output in safetensors?
the exemple images are completely different between checkpoint version... like i love the minimalist dot eye image of the kwen images, but i want that on anima, what prompt did you use?
Anima is biased toward an anime/manga aesthetic. If you want the minimalist dot eye you should move toward another model that has a better style variety, like Krea.