Nova Animal XL
A realistic furry checkpoint model that can have great details on any type of furs, scales and feathers, which aims to be XL-lized Persika 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
Steps: 20-40
(Pony)
CFG Scale: 6-7
Prompt: score_9, score_8_up, score_7_up, BREAK, furry, anthro, detailed face eyes and fur, {Animal Type}
Negative Prompt: score_6,score_5,score_4, source_anime,source_cartoon, 2d, art, cartoon, 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: 3-6
Prompt: masterpiece, best quality, amazing quality, very aesthetic, high resolution, ultra-detailed, absurdres, scenery, photo, realistic, furry, anthro, {Prompt}, BREAK, depth of field, detailed fluffy fur, photorealistic details
Negative Prompt: smooth skin, human, multiple tails, 2d, art, old, oldest, cartoon, graphic, text, painting, crayon, graphite, abstract, glitch, deformed, mutated, ugly, disfigured, long body, lowres, bad anatomy, bad hands, missing fingers, extra digits, fewer digits, very displeasing, (worst quality, bad quality:1.2), bad anatomy, sketch, jpeg artifacts, signature, watermark, username, simple background, conjoined, bad ai-generated
Description
Enhance realism and facial structures
NoobAI EPS v1.1 + Illustrious v2.0 stable DARE applied
FAQ
Comments (30)
👍❓👎Well, let's get started. 5B
OK. The Model 5 is pretty special. The 5A has a strong anime bias, the 5B model is between 5A and 4B and is a more interesting example. After conducting various tests, I can say that the model 5B has a different type of coat, It is more sharp, and the image has increased contrast.
BUT
🤔This doesn't mean that I didn't like this model. Simple tests with basic image generation did not give me a clear understanding. In this case, I will use the 5B as the main model for work. This will help me understand her weaknesses and strengths.
📢I will add my comment later.
I like the checkpoint, but is there a way to prompt for different species? It seems to have a tendency to mix species (e.g. "1 wolf, 1 tiger, duo" tends to end with 2 characters that look like a mix of both species.
Adding "various" or "different species" doesnt do much
Dear friend, use this construction
≫ For duo:
(1girl,anthro,furry, "species", "description of the girl's details and actions"), (1boy, anthro,furry, "species", "description of the boy's details and actions"), male/female, duo, "their common actions"
⋙For trio:
(1girl,anthro,furry, "species", "description of the girl's details and actions"), (2boy, anthro,furry, "species", "description of the boy's details and actions"), group, trio, "their common actions"
⚠️Don't forget to put each character in parentheses.
⚠️If the generation is biased towards one species, then you need to up strength of "weak" species. For example:
🔥(1girl,anthro,furry,(cat:1.2),(Calico:1.1),curvy,big breasts,ass,tail,bottomless,fluffy fur,detailed fluffy fur), (1boy,anthro,furry,dobermann,muscular,male leaning on female,)
A brief opinion (not a truth) of what I tought after one image rendered with my full workflow (5B model) :
First of all, I must say my sceptism when I discover the showcase images of 5B model, especially because it is close at first glance from the 5A model. In comparison, 4B and 4A were more separate each other.
So what's the point judging from the showcase ?
Images of 5B are sharper, undoubtedly. Colors are more vivid (too much ?), and I guess sometimes some flickers artifacts, notably when lighting is a main component of the final image. Speaking of the light, I couldn't say now for sure but I feel like 5B could be currently the best model in terms of "ray-tracing" interpretation. Highlights and shadows are more subtle and give some volume to the image.
After some experimentations, the final image : https://civitai.com/images/84628659
Nothing much different from what I say before. One thing I can add is that the model understand, especially on the inpaint step, what is fur... A BIT TOO MUCH ! I could whine with the difficulty to make fluffy fur with 4B. Here, for those who watch the image, it's the first time I transform so easily a hellhound into a border collie. But it works... And as I feared, there are some flickers artifacts (knee, abs, boobs), but nothing too serious.
So to conclude, hard to say anything more without more testing. It's not bad, it does better than 4B on some points, not everywhere. I think the difference of sharpening will be the hardest thing to manage compared if you're used to the ancient version.
If you've made it this far : Congratulations ! Here's some self promotion : subscribe, follow me, check my work, hit the bell, bla...bla...
And take care.
As I mentioned on my X(Twitter) before, I used the newest feature on MMS t2i called Hires Adjust for the v5B images
I made two versions for each systems as well
One for webUI(A1111 / Forge)
And One for ComfyUI
It can adjust colors, brightness and contrasts for separated highlights, shadows and middles on the pre-hires image and passes it into hires
The 1st image (the one with fox and fire) used 0.9 contrasts on all factors
Non-adjusted images can be found on my X as well
Can we control the color of the images during generation, or is this done only when creating a model?
@MiraKoks This can be done on image generation
t2i >> color adjust >> hires
@Crody Can I give you a little instruction on how to do this? All of this looks like tools for developers, not for ordinary users. Or am I just stupid and didn't understand the obvious things because I work in a Forge, I make promp, press the generate button, then hires. I don't understand where I can find these color adjustments. If this is clear to everyone except me, then you can just ignore this comment. Or give me a guide to better understand how it works. Please
@MiraKoks This can be found on my own model merge scripter which can make the notebook for Kaggle
https://civitai.com/articles/14594/how-to-use-t2i-for-sdxl-without-gpu-or-kaggle-guide
This isn’t the tool for forge or any other webui (at least not yet)
@Crody Thank you for your reply
@Crody Since I generate on my PC, I completely ignored all other tools😓
I've seen the images you mention on your X account. Now I understand better. It's my mistake on that point.
As I said, my opinion on the model is mainly a feeling for now. I haven't used it enough to properly evaluate it. But the add of sharpness is very welcomed to finish on a positive note.
@MiraKoks I'll make the one for Forge/A1111 as extension
And ComfyUI as well
@Crody I don't if you have contacts with Onoma Ai but any news on an open source release of illustrious v3.5? I saw they hit their "stardust" objective so...
@tris99 Sorry I don't have connection with them but I'm looking forward to the newer versions
@Crody Oh, great, that's brilliant💎
@MiraKoks So, I made two versions
One for webUI(A1111 / Forge)
And One for ComfyUI
I'm just no good at this. I keep trying to make images like Disney's Maid Marian, one of my fave cute characters from animation, but I'm no good at this. I'm not familiar with all the right tags for these kinds of characters. I am downloading the 5B checkpoint and will try with this and let you know how it goes.
The checkpoint's output, what people have done with it, looks great and the art is amazing. Great Work. Thank You.
Are you using a LoRA for the character?
@IndolentCat Yes. I found I had wired up my workflow wrong. The MODEL wire has to be linked to the ksampler from the OUTPUT MODEL wire of the Wildcard POSITIVE Prompt Node, the one that loads in the <lora:*:1.0> part... and not from a node before it.
A simple error and now I can generate the character using one of the many ILLUSTRIOUS loRAs for Maid Marian out there. My other workflow worked but this Furry one had a wiring problem, now fixed. My mistake.
This checkpoint works great. Thank You.
For some reason whenever i use img2img in Forge the result ends up gettting blurry, using 'blurry' doens't seem to help. happens in both V2 and V5B
The upscaler models are designed specifically to remove the blurriness and sharpen the image. If you were using i2i without an upscale model, you wouldn't be able to get a clear image. The second factor is the image resolution. If you are trying to make a clear and detailed full-length image of a character with a resolution of 768x1152, then this is a bad decision because the pixel density is not high and the potential for increased detail is low. I hope I made myself clear.💜
@MiraKoks thank you, i'll keep it in mind, for some reason this is the only checkpoint i've had this issue on with i2i, mabye i need to tweak the settings more, thanks again
jbsbiome same here. It's like a mild myopia - bearable, but visible, right? It is either the model, loras or prompting issue, not because of resolution or upscaler fault, as img2img is a generative method, like hires fix, - it must add details even from thin air. As a quick solution - after hires fix or img2img upscale just run a quick 1x upscale with 1x-ReFocus.
mdhyena yes its very slight, i started using a sharpening lora to sort of fix. what is 1x-ReFocus, is that in Forge?
jbsbiome sort of a upscaling model, that just sharpens the image and reduces blurriness. Just run 1x upscale with it and it should do the trick. You can grab it from openmodeldb and put it into models/ESRGAN directiory of your Forge installation folder for it to appear in the list of upscalers at Extras tab.
I have to admit this model is impressive, but I’ve noticed it tends to excessively emphasize fur, often rendering it with a hard, greasy, and wet texture. All samplers—except when using the LCM scheduler, which completely erases fur details—produce the same effect. Additionally, the model seems to struggle with representing the bright feeling of direct sunlight; even with strong prompts, the generated sunny scenes always feel gloomy and overcast.
Try using a LORA like Realistic Furry or AI Slopslayer, also possible your CFG is too high.
furproxy Thanks! i'll try.