Tōru Takagami (高上 透, Takagami Tōru) is one of the main protagonists of Wagaya no Oinari-sama. series.
He is the youngest son of the Takagami family. He is eleven-years-old and has been told by his grandmother that he has strong Yin in his blood and that Tōru may be sought after by spirits and yokai. (source)
I randomly stumbled on him and decided to create Lora of him.
You need to use appropriate age tag.
Trained using Illustrious v2.0
Description
Comments (8)
how much pictures u use for train its shit......
I'm sorry to hear you feel that way. I had 57 images in the dataset. However, the problem is cause it's an old anime the source material has resolution of 704x400 pixels, which is quite a small number to work with compare to like 1920x1080 for 1080p which is standard nowadays (that's 7 times less pixels to work with, even when you consider character is not taking up the whole screen the whole time). I shall research how to improve my process for old anime like this one so I can meet your expectations. Thank you for your feedback as it helps me improve.
@aanonimouss21 as a tipp use Novel AI for creating you can use old Anime chars as preference and it exact make a picture with that character... with that you can make many poses and you can use the pictures for make Lora here :)
@soccerworldai319 Oh, didn't know that. I will check it out. I heard about Novel AI but I never used it before. Thank you <3
@aanonimouss21 Quite the level-headed response to (in my opinion) unfair criticism. I really respect it, a lot.
I love your content, and appreciate everything you do to provide shot-her character enjoyers like me with content.
As for training, I'm not sure how much you know, but... people use consistently cropped images across all of their dataset; regardless of whether they got the images from a 480p show or a 1080p show (512x512 is generally the standard). And some go above and beyond with upscaling those images, though that's a lot of effort if you consider this just a hobby.
Some tips:
- use https://fancaps.net/ to see if it has your show. The screenshots on there are usually slightly higher quality than those from yohoho sites, and it has screenshots of every frame of the show they list. Then use https://cropper.launchigniter.com/en to quickly make those images the same size (again, something like 512x512, though you can play around to see what appeals to you).
- Get yourself https://github.com/Bionus/imgbrd-grabber for quickly finding images of characters across multiple Booru sites to save a lot of time (mostly fanart)
- Try to avoid using CivitaI's tagger, as it's pretty notoriously terrible. I recommend using something like https://github.com/jhc13/taggui
Either way, I'll always love what you put out.
Also, do you ever plan to add a link for commissions? Doing a load of them will definitely improve your abilities.
@Your_Friend_ Thank you for your kind words. I don't let negativity get to me and I try to better myself every day :). Negative feedback can give you a lot of information on what you can improve!
Thank you for the tips, I like to get the source material (like anime for example) and manually take screencaps of it. That way I can get best quality of images. It does take a bit longer to do that but its usually worth it.
What I learned was that its more important to keep the aspect ratio than specific resolutions, so I keep crops at 3:4, 4:3 or 1:1 ratios, and to get it all done properly its all manual work.
That imgbdr-grabber looks great, I was only checking Danbooru but that one is quite limited unless you pay for it so I will give this one a try.
I train on tensor.art and the tagger there is good, though UI is not perfect when you need to clean up the tags so I will give taggui a try!
I probably should work on getting the word out that I do take commissions, but I'm unsure what is the best way to do that :/ do you have a suggestion for that? Maybe I just need to advertise my Ko-fi page.
@aanonimouss21 Damn, I have to appreciate you putting in the manual labour to ensure quality. That's already above what most LoRA creators do, lol.
Yeah, Danbooru was never good. Gelbooru is the way to go with the "display all site content" activated in the settings. chan.sankakucomplex.com with a free account and the filter disabled is king though for... people like me and yourself. But either way, that imgbrd thing I sent should cover your needs for LoRAs (can't recall if it has sankakucomplex as an option though, but that stuff is a bit hardcore anyway, I'd imagine).
Ko-fi is a good suggestion for payments. As is Fiverr. Though, Fiverr has slightly higher fees than Ko-fi. I believe the lowest fees method is taking payments directly via PayPal. But, from what I recall, it's not possible to hide your name on PayPal, so anonymous payments aren't really possible.
There's also the yikes method of accepting gold buzz, but, as everyone knows, gold buzz is pretty much completely worthless on this site these days. It might be an okay option if you want to accept low payments for your work while you gain confidence though. But yeah, it's pretty much a nothing burger.
@Your_Friend_ <3 I will probably create some pricing chart and start from there, I'm probably gonna go with lower prices till I get going. This is not my main source of income anyway, but I enjoy doing it and happy others like it too. Probably gonna keep Ko-fi and maybe start Fiver if i see need. But yea as you said buzz does not do much for me, especially with this recent change I can't even convert it to real money anymore (though I was not yet there with the minimum you need to be at to be allowed to do that).
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.