This model is a merger of fluffyrock and BB95 furry mix.
V1 is a mix of fluffyrock-576-704-832-960-1088-lion-low-lr-e19-offset-noise-e4 and bb95FurryMix_v50
V2 is a mix of fluffyrock-576-704-832-960-1088-lion-low-lr-e27 and bb95FurryMix_v70
V3 is a mix of fluffyrock-3m-e77-offset-noise-e62 and BB95furrymix_v90
V6 is a mix of fluffyrock_3mE86OffsetNoiseE71 and bb95FurryMix_v110
V7 is a mix of fluffyrock-576-704-832-960-1088-lion-low-lr-e92-terminal-snr-e65 and bb95FurryMix_v12
I recommend adding "simple background" to the negative prompt.
Works best with e621 tags, use spaces no underscores,
Works well in a variety of sizes and aspect ratios up to 1088x1088, does widescreen really well.
I normally use DMP++ 2M, 50 steps, CFG 11, 776 to 1088 or 960x400 for widescreen. If I do hires fix I use R-ESRGAN 4x+ Anime6b upscaler, 0 steps, strength 2-5, upscale by 2.
Works well for anthro and feral ok for humans, sfw and nsfw, a lot of creatures and pokemon, and with any LORA I have tried with it.
Can do a range of styles from super realistic to cartoony and photo backgrounds to simple ones.
You can find more images from the model here. https://e6ai.net/posts?tags=furryrock_%28model%29
Description
version 3.0 is a mix fluffyrock 3m-e77-offset-noise-e62 and BB95 furry mix - 9.0
FAQ
Comments (20)
Why are the v3 images all canine
Because I prefer canines, but v3 is actually the only one that i didn't put all canine there's a cheetah.
@firewolf890聽Since it's a set of showcase images it would be nice to see what else it can do
Can I ask what embeddings you may have used? I can't recreate any of the samples.
The only thing i can think of is the CFG rescaler, I got it to test out fluffyrock's vpred model and it might have still been on.
@firewolf890聽Mine won't do the hair or shading anything like the example shots. Is the CFG rescaler an addon? I wish I could post an example of my results. They're nothing like the samples, but I can't figure out why.
@squidboimusic it might be something else then the rescaler mostly effects the background for me but here's a link to it https://github.com/Seshelle/CFG_Rescale_webui
@firewolf890聽Thanks for the link! I'll try downloading it, and if I still can't reproduce the examples. I'll try to post one of the results that I get! 馃憤
@squidboimusic聽If you use Discord you can join the Furry Diffusion server https://discord.gg/wcyyBz6G or message me firewolf890. I'll be more likely to see your message there.
It is definitely due to CFG Rescale, I have no idea why this model requires it while others don't, but you should really put that in the info for this model. Without it, everything has basically no detail and is unusable, but with it, it's really good.
I downloaded the CFG Rescale add-on, but I still can't remake the samples for some reason. I wish I knew what the CFG Rescale was set to on each of them.
@squidboimusic聽I didn't use the cfg resaler on any of these but most of them do have inpainting done to them.
@upscaleanon537 a lot of people have used this model without it and I didn't use it for any of these.
@firewolf890聽I friended you on Discord. I'll show you my results. I can't seem to remake the samples at all. But I'm not even getting anywhere close.
@firewolf890聽You sure they actually use this? And the V6 version? Maybe your V3 version is different. No matter what I did, it would not yield any good result without CFG Rescale.
Also you literally said earlier that it might be due to CFG Rescale, make up your mind lol.
If anyone has downloaded CFG Rescale and is still struggling, try changing clip skip from 1 to 2, or 2 to 1.
@upscaleanon537聽lol ya if i didnt use it it wouldn't come up as the model in the meda data and I made up my mind when I removed cfg rescaler and haven't gotten any different results from before I dont use it. What's probably going on is I remove the artist tags from the prompt if I'm inpainting out a watermark, that also changes the seed from the original generation in the process so you will not get the same image. If you are just copying the prompt from here make sure it has some artist in the prompt if not try adding the artist part of the prompt back in from one that still has it.
Dumb question, but how does one go about finetuning this model?
This model doesnt seem to have female perspective, female pov, makes me sad :c
It depends on e621 tags, check e621 for the kibd of pictures you want to replicate and use the tags of those. Or maybe there isn't enough of those because furries ain't like that :D
Does this model have reptile and avian cloaca and arthropod abdomen genitalia?