🎞 LoRA is trained purely on anime screencaps
🗝 Activation tags are in the right panel
📸 Always use ADetailer to get detailed eyes
🛠 Technical notes:
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 (13)
😍 👍🏻
The model can be improved but not bad, if you add the description of the character you can do it.. although I must say that sometimes the face like the eyes, nose and mouth come out strange, good work
Wdym come out strange? What you mentioned is mostly the issue on checkpoints, not the LoRA itself.
@LittleJelly I can assure you that even when changing checkpoints sometimes the face has a "real" trait unless you write something in the prompts or something else... it doesn't always do it.. but it does. then things happen to everyone based on what they do... mine wasn't a criticism anyway
@Hitoshirp I'm not taking it as criticism, I'm trying to figure out what's wrong. So far you didn't give me exact answer. Like "write something in the prompts" doesn't really say much.
@LittleJelly I don't know what's wrong and that sometimes when you change the seeds or cfg saling... the character's face is no longer an anime face or rather... it has the nose, mouth and eyes of a real character... maybe it's just a mistake but I noticed it if you keep especially if you keep 1 alpha... I solved this problem by putting it even just at 0.9 alpha.. this is the problem
@Hitoshirp I just gave this LoRA another few tests, with 3 different checkpoints, another 3 different CFGs (4, 5, 6), LoRA strength at 0.8 - 1.0 and never encountered anything about her face being weird 🤔 All I can tell you is that checkpoints based on Illu 0.1 and higher, have a bit burnt some tags and it might cause that issue. However as I said, no issues found on my end with default tags I provided.
@LittleJelly my checkpoint is for illu.. but it could be because it's a recent checkpoint... thanks anyway
In my testing, only thing I would add is "Seiko Ayase" and "(hair up:1.2)" because I would get loosen hair when she is sitting, totally random
don't add names to the prompt... idk who gave you that idea but it's wrong when you are using a custom lora.
I haven't had any issue with her hair so it might be a checkpoint issue.
@LittleJelly fair enough for checkpoint part, I wouldn't even consider that as Lora issue. But now you got me curious why no name in Lora prompts? I see a lot of people doing it? Thank you for answering in advance!
@probrodzoni509 People who do it most likely learned from old guides. And also a lot of online trainers offer to have a general tag and people use it without knowing when it's good and when it's bad. Especially with multi outfit charas, having a general tag is a bad idea as you never know what it will absorb and bleeding can be expected in the end result.
I explain it in my guide a bit more in-depth.
@LittleJelly well thank you for this short explanation :)
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.