YesNoOk
avatar

Nishimura-styled CvS portraits w/ Stable Diffusion! (Read 213856 times)

Started by Dawn de Era, July 17, 2023, 06:59:23 am
Nishimura-styled CvS portraits w/ Stable Diffusion!
#1  July 17, 2023, 06:59:23 am
  • ***
  • Come On You Know LGx
  • biblical (yea)
    • USA
    • Skype - eddsworldtrash
    • remixsparta.wix.com/mugensdawn
I've been making a ton of these for a while now, may as well dump some!

Most of these are generated using Stable Diffusion and manually cleaned in post. I use the NovelAI DIffusion base model and a Nishimura style LoRA that I trained myself. alongside character LoRAs that I either train myself or find scavenging on CivitAI.

Cleaned images/WIP cleanings:

Spoiler: Cleaned/WIP cleaning (click to see content)
(Hisomi and onwards is when I started using the Stable Diffusion upscaler.)

Unclean generations:

Spoiler: Unclean (click to see content)

Feel free to use these for anything you like, though I suggest waiting 'til I clean them up first, lol.
Last Edit: December 11, 2023, 09:51:55 am by Dawn de Era
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
#2  July 17, 2023, 08:30:50 am
  • *****
  • A.K.A. NED
  • I like to draw fighting game characters...
This is pretty interesting !
I'm pretty noobish to that. Is that a paid service ? I would be curious to test that.

-What is the exact prompt you typed for the maid character?
I want to understand a bit.

Also did you had to add some specific pictures to help them understanding Nishimura style or the character appearance or just prompt ?

Thnaks.
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
#3  July 17, 2023, 09:07:08 am
  • ***
  • Come On You Know LGx
  • biblical (yea)
    • USA
    • Skype - eddsworldtrash
    • remixsparta.wix.com/mugensdawn
I'm pretty noobish to that. Is that a paid service ? I would be curious to test that.
Nope! If you have a powerful enough GPU, Stable Diffusion is open source and completely free! There are paid online alternatives for lower end PCs, though I know nothing about them.

-What is the exact prompt you typed for the maid character?
I want to understand a bit.
That Marie portrait was helped massively by the character LoRA, so I'm not sure how you would generate it with just the base model. With an anime base model, though, I believe
Code:
maid apron / waist apron, maid headdress, mob cap, frills, puffy sleeves
are the Danbooru tags for the outfit.

Also did you had to add some specific pictures to help them understanding Nishimura style or the character appearance or just prompt ?
I trained a style LoRA on every Nishimura CvS1 and CvS2 portrait. That is what is responsible for the style replication.
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
#4  July 17, 2023, 11:43:15 am
  • ****
  • Pixels are atom's of resolution,Low-res or Hi-res
    • Turkey
    • metekervan26@gmail.com
Is there any chance we can have the trained model or sth
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
#5  July 17, 2023, 12:03:10 pm
  • *****
  • A.K.A. NED
  • I like to draw fighting game characters...
I'm pretty noobish to that. Is that a paid service ? I would be curious to test that.
Nope! If you have a powerful enough GPU, Stable Diffusion is open source and completely free! There are paid online alternatives for lower end PCs, though I know nothing about them.

-What is the exact prompt you typed for the maid character?
I want to understand a bit.
That Marie portrait was helped massively by the character LoRA, so I'm not sure how you would generate it with just the base model. With an anime base model, though, I believe
Code:
maid apron / waist apron, maid headdress, mob cap, frills, puffy sleeves
are the Danbooru tags for the outfit.

Also did you had to add some specific pictures to help them understanding Nishimura style or the character appearance or just prompt ?
I trained a style LoRA on every Nishimura CvS1 and CvS2 portrait. That is what is responsible for the style replication.

Thanks for the reply !
Sounds very interesting but Also very complicated. At least for me.

Do you have a link to the place to DL this free version ?
I want to make some tests with my pc performance

Also, not sure I got the LoRA thing.


And as Burningsoul said.
Would be great to have access to this trained AI.
I'm not even sure it can be exported, shared. But I would like to try.
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
#6  July 17, 2023, 04:49:48 pm
  • ***
  • Come On You Know LGx
  • biblical (yea)
    • USA
    • Skype - eddsworldtrash
    • remixsparta.wix.com/mugensdawn
To use Stable Diffusion, I use AUTOMATIC1111's WebUI. Check the system requirements before giving it a shot, though.

As for releasing the "trained AI", I assume you mean the Nishimura style LoRA itself. I'll see if I can upload it to CivitAI sometime.
Using A.I. to change a character's sprite style? (Re: Nishimura-styled CvS[...])
#7  July 20, 2023, 09:52:12 pm
  • ****
    • Brazil
    • chng.it/QbGdS8rbnh
Woah, all those results are surely impressive...! :datass:

Spoiler: Cleaned/WIP cleaning (click to see content)

Spoiler: Unclean (click to see content)

Sir, now you're giving me some wacky ideas... :S

If there's ever a portrait of BF in this style as well, then I'll naturally feel obligated to put those ideas into practice... Just saying. :innocent:



Sorry if I'm being off-topic from now on, but since we're talking about A.I., I feel like this might be a proper place to ponder about something -

Spoiler, click to toggle visibilty
Re: Using A.I. to change a character's sprite style? (Re: Nishimura-styled CvS[...])
#8  July 20, 2023, 10:49:31 pm
  • ***
  • Come On You Know LGx
  • biblical (yea)
    • USA
    • Skype - eddsworldtrash
    • remixsparta.wix.com/mugensdawn
If there's ever a portrait of BF in this style as well, then I'll naturally feel obligated to put those ideas into practice... Just saying. :innocent:

God forbid.


Sorry if I'm being off-topic from now on, but since we're talking about A.I., I feel like this might be a proper place to ponder about something -

Spoiler, click to toggle visibilty

That would probably be a nightmare to get working reliably. Stable Diffusion really does not play well with pixel art in what I've observed, but I suppose what you wanted could be achieved with img2img and a style/concept LoRA. The result would probably look more like mixels, but I guess the theory isn't that far from what can be tried.
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
#9  August 01, 2023, 05:37:45 am
  • ****
  • Rurouni Fan
  • The dream is nothing yet drive us to the end.
    • www.youtube.com/playlist?list=PLaWlQGjKyJWoXxMR8nrea7P4R9C0JnA-_
If you have a powerful enough GPU...
Ha! Not for me.  :P
I would love to see characters from Rurouni Kenshin and Inuyasha in this style.

Anyway, I really like Nishimura's art in CVS2 so seeing new renders in a style close to that one is pretty cool, even if they need some retouching. Speaking of which, I hope this isn't a problem:


I admit that I rushed a little but I wanted to finish it quickly. Again, if not an issue I would work on 1 or 2 more. I hope you can continue to share more of these results with the community.
One day someone will rip the Inuyasha stages... Not today nor tomorrow, probably not even in this decade. But you know, someday... :')


Did you play the new
Spoiler, click to toggle visibilty
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
#10  August 01, 2023, 11:50:19 pm
  • ***
  • Come On You Know LGx
  • biblical (yea)
    • USA
    • Skype - eddsworldtrash
    • remixsparta.wix.com/mugensdawn
Again, if not an issue I would work on 1 or 2 more.

Go right ahead! I haven't time to get off my ass and do them myself anyway. Hell, haven't even had the time to do any more generations, but rest assured I'll be doing more. I need a new portrait for Kyanta 2 Rogue as well, and I've already got her character LoRA ready..
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
New #11  August 02, 2023, 04:08:00 am
  • ****
  • Rurouni Fan
  • The dream is nothing yet drive us to the end.
    • www.youtube.com/playlist?list=PLaWlQGjKyJWoXxMR8nrea7P4R9C0JnA-_
Nice. So here they are:

Spoiler, click to toggle visibilty
One day someone will rip the Inuyasha stages... Not today nor tomorrow, probably not even in this decade. But you know, someday... :')


Did you play the new
Spoiler, click to toggle visibilty
Last Edit: December 12, 2023, 02:51:29 am by Slayer.
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
#12  August 09, 2023, 07:15:15 am
  • ***
  • Come On You Know LGx
  • biblical (yea)
    • USA
    • Skype - eddsworldtrash
    • remixsparta.wix.com/mugensdawn
Didn't have the time to try and generate any of these in a good bit, but I finally got around to doing one for Rogue (Kyanta 2). This is cleaned to the best of my abilities.
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
#13  August 14, 2023, 05:51:33 pm
  • ***
  • Come On You Know LGx
  • biblical (yea)
    • USA
    • Skype - eddsworldtrash
    • remixsparta.wix.com/mugensdawn
Someone else asked me to do these, and I managed to easily nab some character LoRAs, so why not? Upscaler was a bit of a pain to hassle with though.



Neither were manually touched up.
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
#14  August 15, 2023, 03:34:20 am
  • *****
  • Legend now but still learning
    • USA
    • http://cvsnb.deviantart.com/
is your going upload the lora online? these are pretty cool
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
New #15  August 18, 2023, 01:40:58 am
  • ****
  • Rurouni Fan
  • The dream is nothing yet drive us to the end.
    • www.youtube.com/playlist?list=PLaWlQGjKyJWoXxMR8nrea7P4R9C0JnA-_
To be honest, when I first saw the original post, I wanted to edit all 7 of the original pieces you posted. So now they are "finished":

Spoiler, click to toggle visibilty
One day someone will rip the Inuyasha stages... Not today nor tomorrow, probably not even in this decade. But you know, someday... :')


Did you play the new
Spoiler, click to toggle visibilty
Last Edit: December 12, 2023, 02:49:48 am by Slayer.
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
#16  August 18, 2023, 10:32:26 am
  • ****
  • Was it something I said?
Toga and Uraraka look so good!!
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
#17  August 18, 2023, 08:02:36 pm
  • ***
  • Come On You Know LGx
  • biblical (yea)
    • USA
    • Skype - eddsworldtrash
    • remixsparta.wix.com/mugensdawn
Spoiler: Hisomi is hiding in here (click to see content)

Damn! I just might use this for my Hisomi update! (Though it still needs a few little things, especially with Mega-chan's claw things)
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
#18  October 02, 2023, 01:53:27 pm
  • ***
  • Come On You Know LGx
  • biblical (yea)
    • USA
    • Skype - eddsworldtrash
    • remixsparta.wix.com/mugensdawn


Vanessa (Five Nights at Freddy's: Security Breach), WIP cleaning

I have a feeling she should be a grappler.
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
#19  December 11, 2023, 09:54:51 am
  • ***
  • Come On You Know LGx
  • biblical (yea)
    • USA
    • Skype - eddsworldtrash
    • remixsparta.wix.com/mugensdawn


I've been learning a lot more about Stable Diffusion while I've been gone. Elirin in particular is when I started using inpainting for fixing things, and I almost think she doesn't need any cleaning!
Re: Nishimura-styled CvS portraits w/ Stable Diffusion!
#20  December 11, 2023, 02:53:24 pm
  • *****
    • USA
    • www.trinitymugen.net
Those are quite impressive. The more you tinker, the better the result. They could pass the initial sight when paired with official work. I may look more into this myself to get chars that weren't in CvS.