Anima version is available!
Nova Furry XL
A 2d/2.5d furry checkpoint model that can have great details on any type of furs, scales and feathers, which aims to be XL-fied Peaki Furry
Rules
You cannot use the generated images for commercial use if it's not edited (or just turning it to black and white)
Advertising is always welcome
Recommended settings:
Sampler: Euler a for best performance
Steps: 20-50 steps
(Pony)
CFG Scale: 6-7
Prompt: score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, source_furry, BREAK, furry, anthro, detailed face eyes and fur, {Animal Type}
Negative Prompt: blurry, jpeg artifacts, username, watermark, signature, normal quality, worst quality, large head, low quality, text, error, missing fingers, extra digits, fewer digits, bad eye
(Illustrious)
CFG Scale: 3-5
Prompt: masterpiece, best quality, amazing quality, very aesthetic, high resolution, ultra-detailed, absurdres, newest, scenery, furry, anthro, {Prompt}, BREAK, depth of field, detailed fluffy fur, volumetric lighting
Negative Prompt: human, multiple tails, modern, recent, old, oldest, graphic, cartoon, text, painting, crayon, graphite, abstract, glitch, deformed, mutated, ugly, disfigured, long body, lowres, bad anatomy, bad hands, missing fingers, extra digits, fewer digits, cropped, very displeasing, (worst quality, bad quality:1.2), bad anatomy, sketch, jpeg artifacts, signature, watermark, username, simple background, conjoined, bad ai-generated
Description
CLIP-L Fix and controlling the details
NoobAI EPS v1.1 + Illustrious v2.0-stable DARE applied
FAQ
Comments (57)
Hi! What was updated to v12 if I may ask?
CLIP-L corruption fixing and some enhancing on details and posing
@Crody Aight! Thank you. Yeah, I've been noticing some better gens
This one's a lot less noisy than v11. gj.
Any ETA for adding it to pixAI?
Their system for implementing models have some issues currently and I cannot import them
After they fix it, I'll post new versions to PixAI
@Crody 🥰👍
@Crody Can't you just upload them to your profile manually?
@123inactace I don't know why but manual uploading a model takes more time than on CivitAI and sometimes the file gets corrupted
@Crody I think it'll be a while before they'll fix it. Can you try uploading it manually, please?
It took me a few tries to figure out that the best scheduler to avoid images with artifacts was Beta when using Euler Ancestral. Has anyone else used Beta for sampling? I'm curious to know what others are using, especially since it's usually not available to tag for uploaded images.
I'm using euler_ancestral normal for base generation and euler_ancestral simple for "high-res" in my generations. At least in comfyui I'm getting decent results with it.
Hey there! Just a question, how do I select version 12 do generate images? I’m only able to select version 11 or version 6. Thanks for the info.
Wait until 0:00 AM GMT
@Crody the wait is killin me, lol
v12 is awesome. Probably your best one.
Honestly no idea how you create pics without watermark. No matter how I promt (Even if I use the same neg like u in ur example pics) I get watermark artifacts in like 90% of the pics
If the prompt have non-proper tags (which is not listed on danbooru, r34 or e621), they tend to appear on images
Artist tags also do this so for those cases, adding (text, watermark, signature, artist_name, logo:1.5) to the negative prompt would prevent watermark issues slightly better
@Crody Thanks a lot! Btw your work is awesome. Best furry model out there!
I'm staying with version 10, this version still has the same problem as version 11 with very noisy images.
Did you use the guides on the article?
https://civitai.com/articles/19107/crodys-illustrious-noobai-image-generation-tips
If you're using DPM++, score tags or CFG above 6, it'll have lots of noises
I'd suggest using Euler A and CFG around 5
Since I switched to V12, I have a lot of issues with low quality and bad hands.
I have the same issue on most of this collection models recent versions. (this one, anime, reality and animal)
what vae should i use for this?
VAE is already baked
@Crody thank you
Why do I get this error? I'm using Easy Diffusion 3.0. I haven't installed anything extra. I don't understand what to do. Error: Could not load the stable-diffusion model! Reason: 'Embedding' object has no attribute 'wrapped'
When used as a base model for Kohya_ss LoRA training, the v12 model fails to load during "loading text encoders from checkpoint" with the error:
Unexpected key(s) in state_dict: "text_model.embeddings.token_embedding.wrapped.weight"
v11 has worked fine for training. Perhaps this was introduced while trying to fix the Clip-L corruption?
Soooo… since this site is dead now that NSFW isn’t aloud unless you’ve got a fat wallet any update on if this will ever get ported to PixAI? I’d really like to keep using it. But I can’t on this site anymore. (If there’s no reply I’ll just assume the answer is no and never)
Apparently some other user already reuploaded it using manual upload
https://pixai.art/model/1933417134452145535-Nova-Furry-XL-Illustrious-v120
@Crody okay, is that bad? I’d hate to support someone stealing from you.
@Nexious No it's not bad thing
PixAI won't benefit any profit to the users that uploaded the model (like Yellow Buzz system on this site) so doing it doesn't violate my policy
@Crody 🥰❤️ thank you.
@Crody Can you by any chance claim the model to add the negative prompt and the settings, so that by default when someone selects this model it'll put the recommended settings already?
@Nexious I didnt think PixAI, TensorArt or SeaArt allowed NSFW gens at all
@123inactace The best way I can think of is to select v11 first, and then select v12
I know it's not convenient but that'll do it
@LatteLeopard PixAI can do them but only on browser version (which works on iOS and any other platforms as well but no apps)
Click on Profile and select Settings >> Content settings and you'll find the settings to allow them
@LatteLeopard Tensor can still generate NSFW material; the only condition is not to use direct or explicit terms.
That's it, I'm downloading every liked model to my poor laptop. Big props to the creator though, this is honestly the greatest model ever, hope I can find more of their stuff in the future. Just elsewhere, I guess
I’m not happy! That still only buzz yellow
Yeah!
I feel like I've asked this question before so forgive me for sounding dumb, but does v12 use strictly danbooru tags or are e621 and rule34 also supported when prompting? I haven't kept up with the updates for this model and wanted to try it out finally but will support regardless.
Main stream of the prompt would be danbooru but both e621 and r34 tags are supported as well
@Crody Thanks Crody. I appreciate the answer and I like how strong the prompt adherence is. 11/10
Ahm-m... The Best... as always
Good
Short note on Style Tags in Version 12:
They still work, but need a higher weight to be recognizable, usually between 1.2-1.3 will do the trick.
Shout-out to the creator of this model. It is great for comic art in general not just furrys and gives good results from very basic prompts
+ Great with anatomy including hands and even hands holding objects
+ Great for dynamic non-standard poses, including character interactions like hugging, kissing and to some extend even wrestling
+ Knows all common fantasy races like Elves, Orcs and even Draenei from World of Warcraft
+ Can differentiate between different weapon types (Swords, Spears, Axes...) and usually draws them correctly
+ Decent at attribute assignment in pictures with 2 different characters (starts to break down with 3, though)
- tendency to draw watermarks, especially patreon watermarks
- tendency to draw cute little creatures in the background that weren't part of the prompt
- Doesn't work well with style tags
- weak adherence to regional prompter
A lot of the above is true for the Nova Comic XL model but it defaults to a flatter, more basic style and requires more prompt engineering to get great results
Hi Crody, until recently I spent a lot of time using v8B and considered it your best model. But after trying v12.0, I have to say... this is my new favorite! The model’s responsiveness has improved, the poses look better, and the anatomy is excellent. Although there is one downside — I don’t know about others, but I often encounter a “four fingers” issue on the hands, and no negative prompt seems to fix it, only pure randomization helps.
Thanks for the review
4 fingers on hands are known style aesthetic for furry images derived from cartoons
As far as I know, 'fingers' in prompt can prevent some fewer digits issues
And also you can specify the numbers of fingers on each hands by specifying the number of fingers in prompt (for example '3 fingers' for the renamon image)
Check sample images as references
Since the release of v11 and v12, it no longer gives the same quality of realism if I wanted a realism shading quality, as it used to do before on v10 and older versions, wish v10 could have been atleast kept for other models and not looking like a very choppy painting when generating. :(
Since the v11, the coloring / shading gets similar to the IL v3.0: which most of fan base was constructed on
For the next version, I'm planning to make the best of both worlds with the style
Image generation seems very poor in draft mode. Image is very distorted and unrecognizable. Please look into this. Before spending yellow buzz, it would be good to test it in draft mode first.
V12 is great. It keeps up with the current trends for realism models just like everyone else. It works just fine and its more versatile than any of the previous models.
Its like you some of you guys expect him to be restrained by his older models.
It tastes like Furries, it would make even the Joker create an image of Vaporeon for his personal use.
Joker nooo!