Hi all, needed to repost an article to satisfy the 12/14 task for Christmas Walkthrough, decided I'd post some information on how to get results most similar to my pictures for Lily Evendale - Christmas. MODEL TRAINING
I am adjusting this article so that it better fits the task description. I'll hopefully try to describe how I manage to generate images like my loras.
So here is what I do in stepwise fashion when I set up an image to generate.
1) The model I use is Colorburst XL. This is something I've used for a long time now, but I use the pony variant.
2) Next, I look for a lora that fits the appearance of the character I want to make. For example, I use Chifuyu Orimura as a main model, and make my adjustments to it accordingly. This is of course optional as you can try to describe the character yourself, but if you have a very particular image of your character in mind, it helps using a base model to generate consistent images.
3) To give your character a more original look (i.e. adjust to fit your appearance preference), I typically set the lora strength to about 0.5. This helps keep some of the character's original core features but allows you to make modifications that appear more consistently (i.e. longer hair, different color skin tone, different eye color, etc.)
4) Next, I typically use these as my default settings with image generation. It for some reason just works best with this model in my opinion, but feel free to experiment:
5) Now the next part is of course the prompt. As per the model's recommendations, there are some prompt triggers that I typically use:
Positive:
"score_9, score_8_up, score_7_up, score_6_up, source_anime, BREAK 1girl, solo, (ultra quality, UHD, 32k, best quality, dynamic lighting)"
Negative:
"(worst quality, low quality:1.4), (malformed hands:1.4),(poorly drawn hands:1.4),(mutated fingers:1.4),(extra limbs:1.35),(poorly drawn face:1.4), easynegative"
6) Now with all that set, generate! Once you have maybe about 8-10 solid images, you can start model training!
7) Now depending on how much your budget is, if you're a beginner like me getting your feet wet, it's pretty expensive...so if it's your first time, try to use fewer epochs (like 2-3) and maybe even fewer images. It will be good to practice first prior to spending a lot on a high quality image.
8) Also before generating, remember to switch the model theme to custom, then pick Colorburst XL for base model! That will give you the most similar style that I typically use.
9) And with that done, can start the training! The generated images for each epoch won't look too impressive, but the lora actually works better when used to generate. And with that, you're done! Hopefully that helps noobs like me trying to get some practice on this. Enjoy!
