After the Embedding(if you have that) or the pony scores, put this in right after:
(anime coloring, 2d, anime_screencap), source_anime, 2d, original anime style,
Since she was in the same spot the whole time, expect the Magic circle, Praise the magic circle, and Respect the magic circle.
And it will play nice xD
Description
FAQ
Comments (14)
If I'm not mistaken her name was Lolisa in the Dust spinoff and Loliette in the fan translation, thanks for the model mang, hope you continue to make her so it ain't that welded with thr magic circle
Yeahh I may have to crop it out more, but her panties and lower half may not match, but worth a try haha
I know you just came back, but can you also put out images with their metadata info embedded onto them in the future for your LORA releases?
I'm a bit protective of my prompt formula haha xD I think once I make big enough changes, then I can start releasing them haha
@Rule34Diffusion I get that, but even if you don't provide everything in both prompts, I am still curious as to what other settings and values you used for the images such as sampler, checkpoints, number of steps, etc.
@a22681 Yeahh true xD, I could at least do that then, One embarrassing thing I did for months in the parameters was, I been using Clip Skip 1 hahaha up until early this month UI been doing it wrong xD
@a22681 Steps: 50, Sampler: Euler a, CFG scale: 9, Seed: 3891394339, Size: 920x1128, Model hash: 4decfc67f8, Model: My_Anime_CG_Pony-fp16-clip-fix, VAE hash: ab33ad55c6, VAE: sdxl_vae.safetensors,
pretty much the settings used. with Adetailer default settings and models(except the resolution, I go for 640 x 768 for face)
I was avoiding hires fix for a while haha, I make tons of works so I was impatient haha
@a22681the checkpoint is not on here as it's an edit I use for personal use :) The lora model for style is also my own personal one I use.
@Rule34Diffusion Fair enough, at least you're willing to divulge some info, better than none unlike other people which is disappointing.
@Rule34Diffusion I just want to say that while you probably won't provide the other prompt data for a long time if ever, I still greatly respect your decision for it.
I got into a bit of a tiff of another LORA creator earlier who's doing something worse where they obfuscate their prompts by only providing triggerword and LORA weight while just viewing the image metadata on it's own actually shows everything. Scummy move by that person, but that's just how I see it. Some people want to protect the work they put into the prompts and that's understandable, then block it off completely instead of putting only two identifiers in the prompt, but they shouldn't obfuscate and misdirect people.
@a22681 That's a very odd thing to do xD
hell, you can even see the metadata if you open the image on Microsoft Notepad and the top section has everything, I recently discovered that btw haha. So even people without webui or the programs meant for it can do it by default.
@Rule34Diffusion Yeah, the guy had perpetual victim complex and cried plagiarism due to being afraid of his generic style being jacked😂. I've used SDimage.info website as that does a decent job of reading PNG and JPG files especially in batch. But didn't know about Notepad pulling that off too, though it makes sense.
@a22681 haha AI has gotten to a point to having an overall similar look, that's why I think the market is big on Original look of anime/video game/cartoon/cgi animated characters haha xD
@Rule34Diffusion True, I do appreciate the value of mimicking or replicating the source's style as best as possible as I've had a LORA project that I've been working on and off for around a year now in doing it. Though I also do like how people can make their own style while still keeping it within the source material's content (character).
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.