Hey I was just wondering what generators everyone uses. I like mine (NovelAI) but wanted to see what else was out there. NovelAI Doesn't have LORA's and lacks some tools I have seen other generators use. However most of the ones I have seen that offer NSFW content don't generate good content for furry, & feral.

cookie-girl said:
Hey I was just wondering what generators everyone uses. I like mine (NovelAI) but wanted to see what else was out there. NovelAI Doesn't have LORA's and lacks some tools I have seen other generators use. However most of the ones I have seen that offer NSFW content don't generate good content for furry, & feral.

Personally I generate locally on my own GPU with forge WebUI.
But generally speaking Civitai and TensorArt are also popular choices for those who don't have the hardware to gen locally.
Don't know about Tensor, but on Civitai feral is semi-banned. Only MLP and Pokemon ferals are allowed to some extend. That's why it's so good to gen locally, nobody censors you.

ComfyUI on a local computer. The feeling of freedom, and not having to deal with credits, non-privacy, censorship, passwords, etc, feels so good.

Might be slow at times, though

SD/A111 for images. Wan 2.1 for videos. Most seem to be using Comfy, but I'm a Gradio fan and luckily both run with that interface. Loras for both are on Civitai, but not too many are out right now for Wan that are furry tailored.

b1techienne said:
not having to deal with credits, non-privacy, censorship, passwords, etc, feels so good.

Might be slow at times, though

Agreed, it's so nice not having to rely upon Kling now and waste money on bad results. The generation times are rough, but I find it best to generate a batch of 10-15 clips overnight for a project, then check out the results in the morning.

I like Midjourney for the aesthetic power it has, for SFW images anyway.

Meanwhile working with ComfyUI+Stability Matrix for NSFW and local machine stuff. Unfortunately right now it's still a very frustrating experience overall -- you spend a lot of time feeling like you're reinventing the wheel over and over and over again.

ocillet said:
SD/A111 for images. Wan 2.1 for videos. Most seem to be using Comfy, but I'm a Gradio fan and luckily both run with that interface. Loras for both are on Civitai, but not too many are out right now for Wan that are furry tailored.

Agreed, it's so nice not having to rely upon Kling now and waste money on bad results. The generation times are rough, but I find it best to generate a batch of 10-15 clips overnight for a project, then check out the results in the morning.

are video generation times as bad as people say? How long does it take for one video gen?

[/quote]
Don't know about Tensor, but on Civitai feral is semi-banned. Only MLP and Pokemon ferals are allowed to some extend. That's why it's so good to gen locally, nobody censors you.
[/quote]

I actually used to run Tensor Art. For what it's worth I haven't looked back since switching to NovelAI, despite the lack of LORA's Novel runs E621 as their dataset (for their furry model) and it is far more reliable than any LORA I found on tensor. Also Novel just added a new model that allows the ability to make multiple detailed characters in one scene. It works well though there isn't a ton of furry/feral art on that model yet. Enough to get by and do some cool things though for sure.

delken said:
are video generation times as bad as people say? How long does it take for one video gen?

Yes, they are. On my 3060 a video with 81 frames 512x512 Wan14b-480p Q8, will take over a half hour. And a video with the much better 720p model at 720x720 (with the half frame count) will take over 1.5 hours.
Generating video with mid-tier GPU feels like trying to generate images on CPU...
So for stuff that isn't NSFW Kling is still the best option.

ocillet said:
SD/A111 for images. Wan 2.1 for videos. Most seem to be using Comfy, but I'm a Gradio fan and luckily both run with that interface. Loras for both are on Civitai, but not too many are out right now for Wan that are furry tailored.

Agreed, it's so nice not having to rely upon Kling now and waste money on bad results. The generation times are rough, but I find it best to generate a batch of 10-15 clips overnight for a project, then check out the results in the morning.

Wan can actually generate NSFW? I didn't even tried, since it's based on Google's T5, maybe I should give it a try

delken said:
are video generation times as bad as people say? How long does it take for one video gen?

1.5~3h here on a 3060 12GB, getting as close to OOM errors as possible to squeeze in more frames

b1techienne said:
Wan can actually generate NSFW? [...]

Yeah, it can. Nudity is no problem with the base mode alone, but for NSFW actions you need LORA, otherwise the results are poor, but there are quiet a few on Civitai.
Also OOM shoudn't really happen. It is no problem to offload to RAM with Wan.

silvicultor said:

Yeah, it can. Nudity is no problem with the base mode alone, but for NSFW actions you need LORA, otherwise the results are poor, but there are quiet a few on Civitai.
Also OOM shoudn't really happen. It is no problem to offload to RAM with Wan.

I was mostly using HunyuanVideo, but OOM were constantly around the corner, maybe it's because of the way it works... Thanks for the advice, I'll do some tests with LORAs and Wan

I use SeaArt and would recommend it. Once you get NSFW unlocked, you can generate pretty much anything you want with all advanced features available, plenty of models and LORAs to choose from. Plus you receive pretty good amount of free credits that renew daily.

They also offer a video generation but since it's rather a paid option, I haven't tried it and can't tell how well it works.

I use Perchance. Free access, free use, no hassle.

delken said:
are video generation times as bad as people say? How long does it take for one video gen?

For me with a 3080 ti, it's between 15-30 minutes, but a lot depends on the duration and settings you use. There's various optimizers which you can enable that do speed up things, but at the sacrifice of quality/good results. Apparently you can just rent beefy GPU processing online and stream what you want to render to it for a couple dollars an hour (though I prefer to do as much as I can locally).

Given that vid gen just came out for us normies to use, the tech is still very much in the early phases. It actually feels similar to how local image generating felt a couple years ago...
But, in time, it will no doubt be optimized and fine tuned with mountains of LORAs like what we have now for images.

Hi Cookie-Girl,

I use comfy-ui. It's great, because you can do a lot, even with a pretty bad GPU (all of my generations were done using a Mobile 2060 RTX with 4 GB of vRam). So even if you don't have an ultra-mega-super GPU, you can still do some decent work. That said, you should at least have some kind of GPU, otherwise generations are really painful.

If you want to give it a go, I'm sure some of the people here will give you a hand. Worst case even I can help out (but yeah: I'm not a role model here, just playing around as well xD )

'

SD and Automatic1111. Tried comfy, but for now I can get much better results in A1111. Might change my opinion once I figure out the trick to using comfy.

cookie-girl said:
Hey I was just wondering what generators everyone uses. I like mine (NovelAI) but wanted to see what else was out there. NovelAI Doesn't have LORA's and lacks some tools I have seen other generators use. However most of the ones I have seen that offer NSFW content don't generate good content for furry, & feral.

I'm currently using Frosting.ai for everything, including NSFW feral stuff
(nsfw furry stuff actually seems to be the most popular category on there, last i checked)

local gens are also allowed, which is nice

IMO no real issues with the free models for the above, but since inpainting & image/prompt strength are only available on paid accounts,
you may have to do some manual editing/prompt wrangling for non-solo/more complex stuff

there's also 0 wait time/daily limit for images -even for free accounts- which amazes me everytime, haha

see also: topic #291