🎞 LoRA is trained purely on anime screencaps
🗝 Activation tags are in the right panel
📝 Note:
Recommended negative: School blazer doesn't really like the purple line. 2-4 gens out of 6 have it. That's the price for not burning it. Training it for longer or with more repeats would risk burning it.
📸 Always use ADetailer to get detailed eyes
🛠 Technical notes:
LoRA shoud be compatible with Illu models & NoobAI. (Not tested.)
LoRA is compatible with various styles - Lower the weight & remove "anime screencap" from the prompt, if you are getting too strong anime style
Hi. res fix used on all sample images:
Upscaler: R-ESRGAN 4x+ Anime6B
Steps: 10-15
Denoising: 0.3
📬Leave a like, comment, review but first of all, ENJOY ;)
Description
FAQ
Comments (8)
Can you do the other characters
Not anytime soon. My To-Do list is rather long at the moment.
@LittleJelly 😢
Could you could train this again one trigger word instead of requiring all the base words? i.e replace these? "short hair, grey hair, parted bangs, forehead, purple eyes, single hair bun, hair bow, purple bow"
Simple answer: No.
Explanation: It's considered bad practice to have one word describe the whole chara. The flexibility of a lora will be lost. Yes, it will work.. but you never know what stuff has been absorbed by the tag.
You can join the discord I have on my profile (Arc an Ciel - it's full of top creators) and ask there about that if you don't believe me.
Thats a shame, I generally use dynamic prompts extension + a script and needing the base features defined makes the lora less flexible for my use case. i.e the face will always be in view even if that was not the intention like when the character is looking towards the background. I would need to have different sets of prompts for the character depending on the composition instead if a single more flerxible tag.
@ld33 it's not really a single more flexible tag... That's just bad practice. You should simply throw out any tags from the prompt that you don't want to be drawn. This is true flexibility. Not that one tag does everything.