First of all, I don't know what I'm doing really. I just trial and errored my way through dozens of variations of so many models I can't recall at this point. So best not to use it in anything commercial I guess
I wanted western looking but not fully comic style look. Its like 90% there where I want it, but now matter what I tried I can't push it further. I can't get rid of the anime eyes on women and can't make the lineart more prominent - it would be perfect then.
And hey, if you decide to give this a thumbs up or even review it, that is awesome. That stuff helps the most
Description
This is probably the final update for this. Since I don't know what I'm doing I can't fix the problems. To be honest I'm more or less satisfied how this looks.
Adding '(black outline:0.4), (thick outlines:0.5)' to the positive and 'white outline, color banding, motion lines, emphasis lines, speed lines' to the negative helps get rid of the annoying double lineart issue. Couldn't get rid of it when merging
FAQ
Comments (20)
Maybe I got just the thing you are looking for:
https://civitai.com/models/151774
And for generell: You can still push a stylized model in a direction with your prompting(Graphic Novel, Comic style, Line Art) and negative prompting(Anime, Manga, Hentai)
Yeah I saw that model when I was experimenting, not sure its in this
@daroth That's my try on the same Idea for a western comic/cartoon style. There is a lot of anime stuff out there but sadly not a lot of western comics.
@TheAiHelper have you tried: https://civitai.com/models/123841/comiccraft ? With the LoRA: https://civitai.com/models/127212/cartooncraft you can make it even less comic-like and more cartoon-like
@victorc25744 those models are awesome. I'm just really into that Jim Lee inspired comic art with the sharp lines and heavy inking. I love your work and I use it a lot ^.^
Gonna experiment with this and report back!
So far looks amazing!
Thanks, thats nice thing to say. If you do report back, thats appreciated
woah this may be the best western looking model out there!! & made by accident trial and error wow dude well done!!!
thats a kind thing to say, thanks. hope you could experiment with this
Hey Daroth ! Congrats and thanks again for this wonderful checkoint! Any plan on doing a SDXL version ? May could you share your training dataset with me so I can train it for SDXL and release it also for the community ;) that would be awesome!
Since you are using a "thick lineart" lora for every single one of your preview pictures, might I suggest that you link said lora? Otherwise it will be impossible to get results looking like yours.
its a mix of two different loras, I'll look for them
@daroth appreciated!
Samples should ALWAYS be made using ONLY the model LoRA's been trained on. (OP) Samples using 2 or more LoRAs don't make sense.
@javbukkake I agree, having downloaded lots of loras that weren't able to do as advertised.
But on the other hand, all the previews are often made using anything from 1 to 10 loras. I don't see that making sense either, because I tend to consider a lora based on all images on it's page.
@javbukkake exactly
Considering you joined a whole month before I did in terms of joining the site - I dont know when you originally started this - but it's amazing, and it's sad you only got a final --- XD I LIKE SEEING MORE UPDATES -- but it's amazing it's to a point you like it! KEEP MAKING MORE COOL STUFF!
This shows great promise. I do enjoy the thicker outlines