Welcome to Inkbunny...
Allowed ratings
To view member-only content, block by keyword, comment or upload, create an account. ( Hide )
A hundred more in happiness and health
« older
Nausamedu
Nausamedu's Gallery (43)

Krea2 and Wulver - Initial testing and documentation

Medium (920px wide max)
Wide - use max window width - scroll to see page ⇅
Fit all of image in window
set default image size: small | medium | wide
Download (new tab)
page 1
page 2
page 3
page 4
page 5
page 6
Workflow - Multi Model Concept Tester
+5
Last in pool
Alright then, so after testing out Anima, i went over to check on Krea2 and Krea2's "more or less" furry aligned model: Wulver (
wasawasa
wasawasa
kept recommending it )

First impressions are... well.
They use natural language to an extent i cannot yet hope to utilize properly. The entire litany of new words and concepts and how the ai reacts to sentences is.... overwhelming af to put it in a mild and feminine way. Maybe im just overthinking it but.... oh boy. Not a fan of natural language rn.

Anw, beside of that, image accuracy and consistency is phenomenal, but this is not your 6-8Gb illustrous model.
It will just about tap out 16gb vram and 32gb main system ram even on the Turbo version which i am using. The original model is even larger.
Once its loaded however, its extremely fast. Can generate on 5-8 steps so its blazing.

I could not properly test it for furry/anthro generation cause it doesnt really understand these terms properly, so i had to re-generate several times and work around it. Especially feline/felid characters were just full blown cats without a hint of human feature, but that might be cause i suck at natural language prompting. The examples show that i got it working after a while tho.

TL;DR:
I already wrote the basic links and documentation for it, so if you want to try it then help yourself to the images and/or the workflow. It has all related links and settings included. Might make a journal page on it later.

Special mention goes to https://civitai.red/images/148444359
The dude was the one who created that first image, however he did not post the prompt for it, so my learning task was to try and recreate it.
Its not a 1:1 but i think i did a decent job and it got me going with the basics.

P.s.: Dont let the fuzzy paintery look of the example images confuse you. I used a lora for that.
You can produce images so sharp and crisp with it that you can cut your fingers on it.

Keywords
female 1,195,336, female/solo 77,144, ai generated 42,183, canid 34,730, werewolf 12,651, ai 12,259, felid 9,717, workflowattached 92, krea2 25, wulver 9
Details
Type: Picture/Pinup
Published: 32 mins ago
Rating: General

MD5 Hash for Page 1... Show Find Identical Posts [?]
Stats
6 views
1 favorite
3 comments

BBCode Tags Show [?]
 
fluffytailedraccoon
24 mins ago
Once you get the hang of natural language prompting, you won't want to go back to SDXL "tag soup." It's one of the reasons I haven't made much porn.

It helps if you have the characters do human activities. Specifying what they're wearing helps push it towards anthro territory. I've found Krea 2 tends to do better furries in "comic book" mode which is what you see a lot of on my page. It also seems to like certain species more than others - rats yes, roosters no.
Nausamedu
16 mins ago
Yeah, thats exactly whats overwhelming me.
I mean you CAN by all means define EVERY last detail of the image you want, but across styles and lighting methods and photography, and color display methods.... I mean holy shit... i will need to start the entire thing by building myself a basic dictionary just like i did with danbooru tags.

It not from scratch but... also kinda close to it. Also not entirely sure yet where to step. We are still in the middle of transitioning towards a post illustrious base model and there are like 5 models out there which could probably displace it on the long run.
Its too soon to try and heavily invest in one unless you got a specific style of your own.

I mean it works perfectly for you because you stuck to one style. And that way it does indeed work perfectly.
The closest thing i have to that is built around a cartoony lora for illustrious which i tend to use for a bit more painty look.

Anyway, enough yapping from me. I cant say anything negative about it aside from how its still a wee little bit unrefined for furry gens. But aside from that tis most certainly a treasure trove of potential.
fluffytailedraccoon
7 mins ago
Yes, it definitely will require knowing about things like architectural styles, body types, clothing, locations, etc. etc. You'll need a dictionary for sure as Krea 2 seems to know a lot. And you'll have to imply certain things - "dark red paint" as opposed to "blood." If it helps at all my prompts have a lot of those keywords in them that you can draw on. Using a prompt enhancer definitely helps too.

I think at this point, though, I've pretty much wrung all I can out of Krea 2 and am looking for something new. How is Wulver? I've heard it's good but haven't tried it yet,
New Comment:
Move reply box to top ↑
Log in or create an account to comment.