I am someone who is more interested in feral work, I noticed some generators completely fail at feral, so given that, on top of me just not having much experience finding generators, I was wondering if any of you knew generators that can handle feral.
(A little extra would be. 1. The cheaper the better, anything free or with limited free uses is helpful. 2. anything that can do female feral on human is nice)
Thanks for any help everyone

somerandomeeveedude said:
I am someone who is more interested in feral work, I noticed some generators completely fail at feral, so given that, on top of me just not having much experience finding generators, I was wondering if any of you knew generators that can handle feral.
(A little extra would be. 1. The cheaper the better, anything free or with limited free uses is helpful. 2. anything that can do female feral on human is nice)
Thanks for any help everyone

I assume you are talking about online services to generate. None of them is really good.
But for your use case probably try TensorArt. I hear they have hardly any rules and the moderation is very weak there anyway.
Apparently it’s possible to do some stuff for free everyday there.
Also please avoid the “flagship” of AI online services which is Civitai. There you will get banned for human on feral.

But if you want truly uncensored AI, you will have to buy a GPU and generate on your own computer. Then nobody can limit what you can do.

I've had a lot of success in various styles with the following models via Frosting's web interface:

• Yiffymix, Fluffyrock (nsfw ok)
• fluffyrock seems much better for anatomically-correct compositions, but yiffymix seems to render art-styles a lot 'better'
• Anything & Kotosmix (sfw only)
• great for eastern-style realistic depictions, but you won't likely get anything furry-specific
• Deliberate (sfw only? never tried otherwise)
• seems to be the only one on this list that consistently recognizes Fine Art artist names/famous painting compositions
• great for combining an artstyle/movement with less human-centric subjects

But if you plan to do anything beyond solo/pov compositions though, you'll probably need a more advanced model + inpainting/editing capability etc like Silvicultor mentioned

silvicultor said:
But if you want truly uncensored AI, you will have to buy a GPU and generate on your own computer. Then nobody can limit what you can do.

If you're using a browser-based GUI, you'll want to use a privacy-focused browser too. I wouldn't trust Edge or Chrome for this -- there's absolutely possibilities those browsers would phone home.

silvicultor said:
I assume you are talking about online services to generate. None of them is really good.
But for your use case probably try TensorArt. I hear they have hardly any rules and the moderation is very weak there anyway.
Apparently it’s possible to do some stuff for free everyday there.
Also please avoid the “flagship” of AI online services which is Civitai. There you will get banned for human on feral.

But if you want truly uncensored AI, you will have to buy a GPU and generate on your own computer. Then nobody can limit what you can do.

What could I use for a M1? locally?

silvicultor said:
I assume you are talking about online services to generate. None of them is really good.
But for your use case probably try TensorArt. I hear they have hardly any rules and the moderation is very weak there anyway.
Apparently it’s possible to do some stuff for free everyday there.
Also please avoid the “flagship” of AI online services which is Civitai. There you will get banned for human on feral.

But if you want truly uncensored AI, you will have to buy a GPU and generate on your own computer. Then nobody can limit what you can do.

Yeah I have just been looking for what I can, my computer isn't super strong (and I'm pretty broke) so I try to find what I can, plus I'm not even 100% sure how to do it locally, interesting enough I didn't even know Civitai had that as an issue, as I had been using it for that and even saw posts of that kind (though I never posted that content). though it's filter is clearly getting stronger so I am defiantly looking to move.

somerandomeeveedude said:
Yeah I have just been looking for what I can, my computer isn't super strong (and I'm pretty broke) so I try to find what I can, plus I'm not even 100% sure how to do it locally, interesting enough I didn't even know Civitai had that as an issue, as I had been using it for that and even saw posts of that kind (though I never posted that content). though it's filter is clearly getting stronger so I am defiantly looking to move.

Yes, especially furry is being targeted by the new rules on Civitai. I hear from many people that they falsely block stuff on the generator that is in fact not ToS violating.
And as for local gen: Your computer doesn't have to be "super strong" to accomplish something. The most important thing is that you have NVIDIA GPU. Even older cards with 4 GB can still be used, it will be pretty slow, tho but probably still more fun than using Civitai on-site generator with all the new restrictions.

hackintosh said:
What could I use for a M1? locally?

It is possible to use Mac for AI generation but it's difficult. I hear that it is great for LLM, because LLM can do well with the concept the Mac uses (it doesn't have "real" VRAM but rather a big shared memory). But for diffusion models Mac is problematic. The lack of CUDA makes the setup difficult and even if you succeed it will be slower than even old NVIDIA GPUs.

bahufaru said:
If you're using a browser-based GUI, you'll want to use a privacy-focused browser too. I wouldn't trust Edge or Chrome for this -- there's absolutely possibilities those browsers would phone home.

Yes, I would never use Chrome or Edge. I only use a privacy focused Firefox-fork.

silvicultor said:
Yes, especially furry is being targeted by the new rules on Civitai. I hear from many people that they falsely block stuff on the generator that is in fact not ToS violating.
And as for local gen: Your computer doesn't have to be "super strong" to accomplish something. The most important thing is that you have NVIDIA GPU. Even older cards with 4 GB can still be used, it will be pretty slow, tho but probably still more fun than using Civitai on-site generator with all the new restrictions.

It is possible to use Mac for AI generation but it's difficult. I hear that it is great for LLM, because LLM can do well with the concept the Mac uses (it doesn't have "real" VRAM but rather a big shared memory). But for diffusion models Mac is problematic. The lack of CUDA makes the setup difficult and even if you succeed it will be slower than even old NVIDIA GPUs.

Yes, I would never use Chrome or Edge. I only use a privacy focused Firefox-fork.

I don't have a major card period, I have my 2 Ghz CPU, 16GB RAM, and 9 GB GPU memory, and that's. So I have to look for whatever options I can find, I can't afford a laptop and PC so I have to have my laptop.

somerandomeeveedude said:
I don't have a major card period, I have my 2 Ghz CPU, 16GB RAM, and 9 GB GPU memory, and that's. So I have to look for whatever options I can find, I can't afford a laptop and PC so I have to have my laptop.

What is the exact laptop model? Ghz doesn't really means anything these days, but with a recent CPU, generating things without upscaling might be doable

b1techienne said:
What is the exact laptop model? Ghz doesn't really means anything these days, but with a recent CPU, generating things without upscaling might be doable

It has an AMD Ryzen 7 5825U with AMD Radeon graphics.

somerandomeeveedude said:
It has an AMD Ryzen 7 5825U with AMD Radeon graphics.

it would take roughly 10 minutes for generating the 512x512 ComfyUI default workflow (using SD1.5), maybe one hour for a 1024x1024 furry picture with a SDXL model (aka most popular furry models), it's a lot but maybe worth trying anyway

b1techienne said:
it would take roughly 10 minutes for generating the 512x512 ComfyUI default workflow (using SD1.5), maybe one hour for a 1024x1024 furry picture with a SDXL model (aka most popular furry models), it's a lot but maybe worth trying anyway

As long as it doesn't come out as being 512x512, it'll be fine.

angry_puppy said:
As long as it doesn't come out as being 512x512, it'll be fine.

Small images: Anything below 768px in either dimension is likely to get deleted.

This is new, I sort of remember not long ago when the limit was 512... Does it applies to video, too?

b1techienne said:
This is new, I sort of remember not long ago when the limit was 512... Does it applies to video, too?

Video as it stands right now generally get treated with a lot more leniency because the technology still needs more time to mature

jelloponies said:
Video as it stands right now generally get treated with a lot more leniency because the technology still needs more time to mature

Thanks you for this information, it's good to know. For locally-generated videos, a 768x768 resolution is barely reachable on affordable consumer hardware, and they usually recommend sticking to resolutions such as 540p or 720p to increase the chances of getting usable results

b1techienne said:
it would take roughly 10 minutes for generating the 512x512 ComfyUI default workflow (using SD1.5), maybe one hour for a 1024x1024 furry picture with a SDXL model (aka most popular furry models), it's a lot but maybe worth trying anyway

Yeah, an hour for one image is quite a while, but I might try it a few times, while staying on the lookout for good online services

b1techienne said:
Thanks you for this information, it's good to know. For locally-generated videos, a 768x768 resolution is barely reachable on affordable consumer hardware, and they usually recommend sticking to resolutions such as 540p or 720p to increase the chances of getting usable results

There's some very cool stuff forthcoming with forced inversion or somesuch, which is WILDLY speeding up how fast 480p video can be generated. Unfortunately it doesn't seem to scale to higher resolutions and is limited for coherence to about 5-10 seconds max length. But it's wild to see nearly real-time video generation out of it on modern cards, and even prior generation cards are still doing video on it about eight times faster than anything like it.