This one is a LoRA trained on my personal artstyle, a second installment of a project I've also done for Anima, this time in the Krea 2 flavor.

The training dataset consisted of 41 high quality monochrome scans and photos of my own pen drawings, some on grid notebook backgrounds, some on plain white paper.

The LoRA was trained on an RTX 3090 using DaSiWa TrainFlow. This is a 1600-step checkpoint. To use it, preface your prompt with "A drawing in the style of Tom Smigla." and then just describe what you want in plain sentences, the same way you'd normally prompt Krea 2. Adding "grid notebook", "white background", or "paper structure" lets you pick the paper background type, and you can also add a notebook binding on the side by directly referencing it in the prompt. There are also three style switches here, and you can activate them by single-word keywords: "chaotic", "minimal", and "unfinished". Try it out!

I ran all of my tests on Krea 2 Raw with a Turbo LoRA at 8 steps and CFG 1.0 (euler, simple), and Krea 2 Turbo should work just the same. A strength of around 0.8-1.0 is where I'd start, but you can also get as far as 1.2 with some interesting effects.

If you like my overall artstyle this LoRA was trained on, you can find prints of some of my handmade originals here!

If you're curious about the training process I wrote a short piece about it here. There you can also find more info about the training settings and the dataset. You can find more of my work on tomsmigla.com.

Feel free to use it however you like. If you make something fun with it, a credit is appreciated!

Obligatory note: This LoRA is a modified version of Krea 2 Raw, and it's not an official Krea product or endorsed by Krea in any way. Krea 2 is licensed under the Krea 2 Community License Agreement (https://cdn.jsdelivr.net/gh/krea-ai/krea-2@db3984fbc6e13b34c0064990fc2d95ac64d00058/assets/hf_samples/LICENSE.pdf), and by downloading or using this LoRA you agree to its terms.