I am admittedly kind of a noob at this, but if you enjoy my humble work, any donations are greatly appreciated: https://ko-fi.com/insult_to_ninjary
Your generosity helps me stay motivated to keep improving and to keep trying new things!
At a Glance:
Base model: Best with illustrated, but you can try others
Use a VAE?: Yes (baked or your choice)
Clip skip: 1 or 2 (try both!)
LoRA weight: 1 - 0.7 (adjust as necessary)
Hires. fix: use non-default upscaler; denoising strength 0.5 - 0.4
Trigger Words:
no trigger - very loose resemblance to my art style, sometimes none at all
pencil sketch - adds more defined pencil sketch lines; dulls/mutes colors; closer resemblance to my art style
monochrome - forces a monochrome color palette; greatly increases resemblance to my art style when used with pencil sketch above
white background - generally gives you a white-ish background if you want that; all my art used in training had no background drawings, so this gives the most authentic results when combined with the two other trigger words above
You may need to weight the trigger words with numerical values or parentheses if the effect is not strong enough. I generally put the trigger words inside 3 sets of parentheses, like so: (((pencil sketch, monochrome, white background))).
Description:
I am not good at drawing. Never have been, never will be. But a while back, I wanted to see what would happen if I just sat down and gave drawing a more serious, methodical, and patient attempt. The end result was a small collection of pencil sketches created over the course of a few months. They are absolutely not good, but also much better than I thought I was capable of!
Flash forward to the age of AI art, I wanted to see how well a LoRA would be able to replicate my drawing "style," so I scanned my sketches with my phone and baked up this safetensor monstrosity!
I was surprised at just how accurate it is to my drawings, especially when using all the trigger words! Now, you won't want to go into using this LoRA with high expectations. If you want really artistic and elegant pencil sketch images, this is not the place for you. But if you're interested in generating things in a really amateurish sketch style, or just messing around, then I hope you can find some value in this LoRA, even if it's just laughing at how bad I must be at drawing in real life, lol
Troubleshooting:
Trigger words are not always necessary, but generally, you're going to want to use at least pencil sketch to force a bit of my "style" into the image. If you want the image in color, only use pencil sketch. If you want a pure, no-color sketch image, add monochrome. Use white background to remove most/all of the background art (this is truest to my original sketches, plus the background can be kind of busy/messy sometimes).
The LoRA can struggle with anatomy at times, but honestly it works a lot better than I thought it would, considering SD1.5's limitations plus the fact that I'm pretty bad at drawing proper anatomy and body proportions, lol. Lowering the LoRA's weight can help with improving anatomy if necessary. Also, I've noticed that clip skip 2 tends to give more pleasant images, but they lose a bit of my style (which makes total sense as to why they look more pleasant, lol).
The LoRA seems to go a bit too heavy on shading. Probably because I also did in my original sketches, but it looks even more pronounced in the LoRA's images. Haven't found a way around that yet, unfortunately.
The LoRA pretty heavily leans NSFW. This is because, as it turns out, drawing booba is easier than drawing clothes, lol. Even so, it's pretty easy to prompt around this if the images you're getting are too saucy. Use your positive and negative prompts wisely!
If you make anything awesome, please consider posting them to the gallery! That goes for all of my LoRAs! I really like seeing what kind of stuff you all are able to make!
If you have any questions, feel free to ask! I will try my best to help! Also, if you have any ideas for future LoRAs, please don't hesitate to share them with me!