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
Make the coloring similar to the Pony versions
FAQ
Comments (54)
Is there any difference from the 3.0 version?
Coloring is much closer to Pony version: more smooth, more 3d-ish
Am I the only one who has down color saturation? In version 4, by about 20%, I usually raised cfg for more saturation, but even having raised the value twice from the original, the colors did not become brighter. There was no such problem before.
Im getting fried images with all the illustrious versions of the checkpoint in A1111, especially V2
Check what kind of setting you use
For example, if CFG is around 6-7, please set it 4.5
A1111 is just kinda dogshit with illustrious in general imo. I switched to Reforge and its working perfectly. (I really recommend switching away from a1111)
@Crody CFG 4 to 5, Sampler: DPM++ 2M SDE, Schedular: Karras, or SGMUniform, steps 20-40
@MrExileWolf Use Euler A Automatic instead
If you're using reforge, fully update it by pulling the latest git and try again.
I recommend switching to reforge A1111 isn't that great and more slower with Reforge is a lot better and fixes a lot of issues A1111 has
@DrHojo123 The strange thing is that the other illustrious models I've used with A1111 work just fine and do not fry images. I'll have to try what Crody recommended first though.
I have been having similar issues but only with specific samplers. I noticed that DPM++ 2M Karras does indeed have noise introduced with illustrious V4 that was not present on the previous releases. Switching to DPM++ SDE Karras (not 2M) fixed the issue as did using Euler A AYS or DPM++ 2M AYS. Overall I far prefer the results using this release even with the introduced instability on that specific sampler.
@whyum4dth0 This is correct there is noise in my gens using DPM++ 2M. I should've added that DPM++ SDE Karras was also my solution.
Really love this model! Great work!
Am I fundamentally doing something wrong in prompting or does this furry model not know what a sheath is?
Change it to different word depending what kind of definition you want to use
1. If you want to use for sword's sheath, use sheathed
2. If you want to separate it from the sword, use unsheathed
3. If partial, use unsheathing
I hope this will help you
@Crody He means animal genitalia sheath. I noticed this as well. If you prompt 'sheath, balls' (like for a nude male or futa character) you'll get a ken doll with testicles holding a literal sword sheath.
@Pyrelinile Probably using condom or colored penis instead will do the work
@Crody Issue still not solved in 6A. I really don't get it why, the dataset should be more than big enough. Canines, felines, many dragons, etc. all have sheaths. Please FIX!
does it need a VAE?
Ill. 4.0 is such an amazing model, thank you👌
I wasn't too keen on the initial Illustrious versions, but v4 has finally won me over. 🥰 That said, the last Pony version is still a good choice as well. 😉
>Make the coloring similar to the Pony versions
I will love you forever
im trying to use the v4 model but it's stubborn as hell for some reason. Doesn't even know what black fur is, generates pic with grey, orange, anything but black fur. I tell it to make the character crouching and it isn't creative at all, doesn't alternate pics of the character kneeling, or on one knee. It's very strict with your prompt, but at the same time, barely follows it...
EDIT: nvm, lowering the CFG scale to 3-5 really helped, it's still iffy, but better at following the prompt :D
i suggest using clip attention multiply if using comfy ui.
q: 1.2
k:1.10
v:0.8
out:1.25
use low cfg between 2.5 ~ 4.5
dpmpp_sde(cpu or gpu)/euler_a generate stable high quality results. use karras for denoiser.
use 10 ~ 40 max steps, using 4k lora is possible and generate a decent image with only 8 steps.
@OnP01nt hmmm id like to follow these steps but I'm using stable diffusion web ui
Illustrious V3 and Illustrious V1 are my absolute favorites of your models. Illustrious V0.8 is great too!
Illustrious V4 has a different look, I do like it alot. But theres just something about those two that really gives Nova Furry its charm and aesthetic.
I hate that I can't use this online, the generations look amazing!
...I should add that I'm just learning Civitai, and feeling somewhat inadequate. Going to a different website to learn new rules there is....intimidating.
Nova Furry XL is one of the best “just works” furry models that consistently delivers good outputs. In my tests, it generated very detailed images at 1516x2048 at 30 samples using both v4 and V3.
Any chance of this ever being available for on-site?
I wish
If civitai expand the space for doing such, yes
@Crody Would love to see it. Love your work, it feels like half the furry LoRAs have been trained on it... and Pix just doesn't have the breadth of LoRAs available. I hear rumors of expanded capacity for base models so I will keep hoping.
Update: Now they are!
Bit of a heads-up on what the model can do with style tags:
https://civitai.com/articles/12081/baked-in-styles-in-the-nova-furry-illustrious-40-checkpoint
Also available as Gallery for the TL:DR folks:
I really like how this looks with a few artist tags I have together, but prompting "1girl" or "1boy" always gives me a furry, even with negative prompts. Is there a good way to stop that?
Probably adding human to the prompt or adding furry, anthro to the negative prompt would do the job
@Crody I got that to work with the negative and some high weights. Thanks
As someone who sometimes does furry, sometimes anime (and sometimes mixes both species together) I never had issues here. Just add human to the positive prompt and anthro, furry to the negative one (and vice versa). Keep CFG lower around 4, and then it should work. Using OCs always help of course, e.g. it will do Nick Wilde as a furry and Zelda as a human, usually ^^
Wonderfully consistent and pleasant to look at, but I'm having issues eliminating motion lines without manually editing them out of every gen that isn't a static pose with this model.
No complaints otherwise. <3
Adding motion lines, speed lines in negative prompt would do the job
Call me stupid, but I actually love these subtle motion lines, was pretty thrilled when I noticed them the first time :D but still good to know how to "disable" them :)
@Rashy I love them too! Sometimes I just want a cleaner look though. My solution was to add motion lines to the negative prompt as Crody suggested, but to eliminate them entirely I needed to go for "sound effects, motion lines, streaks, lines, dynamic effects, speed lines, motion, movement.
What's the difference between the versions?
Hello, I would like to ask for help. I create my designs on a smartphone and I need help because I don't understand all the elements to successfully use this model. Thank you.
What service do you use?
If you use it on PixAI, check out the prompt for the sample images for learning the parameters
I also wrote down the recommended setting in description
@Crody I only use civitai but I tried to remix to set up the Illustration v4 version but I still have a problem telling me that resources are missing
@_Zurg_ Nova models aren't supported for the onsite generator and because of that, the remix function glitches
If you want to use the prompt setting, you can try "Copy All" and fill the information into the generator or PixAI
@Crody unfortunately nothing to do ... I will maybe try on pc later
This is a great model. Well done. I hope it develops even further, while remaining true to its origins.
Many Upsides:
- Very pleasant 2.5D style
- Looks good in 1536x1024, great in 1920x1280
- Good generalist, also excels with humans / anime
- Very solid with body anatomy, less prone to extra fingers/toes
- Very good character recognition, even when I leave away \(series\)
- Duos work great (trios are more tricky, tho)
- Understands a bit of natural language, but could be better (it's okay)
- Great to rework/img2img older renders done with less good-looking Illustrious models
Few Downsides:
- Still could be better with eyes, but upscaling often helps
- Like most other SDXL models it's a mixed bag for macro scenes. It does destruction effects/physics very well, but buildings always look pretty repetitive and grey. Adding "city" or "town" often mixes buildings sizes. Car crushing not really working yet, same for foot/pawprints
- Body liquids like cum or sweat become really excessive, especially when put in (), compared to other models. Maybe it can be toned down in a future update
- High sampler counts = slower render speeds (compared to stuff like ChromaXLMix), but in return the quality is (much) better
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.