- Sep 22, 2019
- 75
- 161
So, the full version was leaked? With premium "Krake" model? Can someone upload it here?
You must be registered to see the links
You must be registered to see the links
Last edited:
- Sep 3, 2016
- 2,020
- 4,050
It is useful only for competition etc. Normal user CAN'T run model bigger than 2.7B because of graphic card's memory limit.
These models need powerful server farm to operate.
- Mar 3, 2021
- 604
- 1,136
From what I've been able to gather, the leak includes their image models which have to be reconfigured to work on something like Stable Diffusion, and possibly some small/incomplete datasets for their text engine? I don't think the whole beefy text model dropped, which would be both emotionally devastating to the dev team and unrunnable for someone getting their hands on it unless they had a few Quadro cards lying around.
Last edited:
- Dec 2, 2016
- 150
- 391
Yeah, no chance in hell you're running this without some beefy cards, like 50k worth of them.
Also, kraken is very much still in development with v2 coming out either later this year or sometime early next year.
Also, kraken is very much still in development with v2 coming out either later this year or sometime early next year.
- Aug 20, 2018
- 3
- 4
For the image gen model, it got leaked so you might be able to find a torrent somewhere, but you'll need some technical skill and a beefy GPU to use it. Though I'm not sure if you could make images as big as the NovelAI site can, and unless they somehow get hacked again or willingly release them, you won't get any improves that they make. If you don't want to use leaked models, you could use either base StableDiffusion or WaifuDiffusion (for anime), but you'll still need a somewhat powerful GPU.
As for the text models, I'm not sure if they got leaked as well, but you could just use
You must be registered to see the links
, which is built for running local models and comes with a few of its own. Again, you'll need a powerful GPU for the larger models, at least if you want a response in a reasonable timeframe. Or you could run it on Google Colab like the good old days, but since you want an offline version that probably isn't an option.TL;DR Yes, but it'll take some work (and you need a powerful GPU)
- Dec 2, 2016
- 150
- 391
For text there is offline variants of this that are based on older models, but it needs at least 12gb of video ram to use it at a decent pace, which is like 5mins per generation.
img based models are about the same I think.
In short, yes there are, but they are vastly inferior models and require a beefy computer to run.
the current production models for both text2img and text generation were leaked, so they're just as good as what's running at novel ai
as for if you can run it offline ... yes and it's been done but practically no unless you have very sharp IT skills and willing to dump 1-2 weeks worth of time into it. the software to run it got leaked too and people have managed to get the inference servers running
on a more 'cost' level though the inference models are ~40 GB ( about the same as the slim base model gpt-neox), so if you want speed (GPUs) it would cost you about 500-2000/mo depending on the exact deployment model if you want to keep to running all the time. you could just run this with CPU if you happen to have a server with 64-128 gb of memory laying around (they're not super expensive, like 100/mo), but it'd be about 20x slower.
mostly elaborating just in case anyone is curious. the pricing of NAI isn't high enough to make it worth it to rebuild it for anything other than fun, plus we should strive to support cool projects like this
as for if you can run it offline ... yes and it's been done but practically no unless you have very sharp IT skills and willing to dump 1-2 weeks worth of time into it. the software to run it got leaked too and people have managed to get the inference servers running
on a more 'cost' level though the inference models are ~40 GB ( about the same as the slim base model gpt-neox), so if you want speed (GPUs) it would cost you about 500-2000/mo depending on the exact deployment model if you want to keep to running all the time. you could just run this with CPU if you happen to have a server with 64-128 gb of memory laying around (they're not super expensive, like 100/mo), but it'd be about 20x slower.
mostly elaborating just in case anyone is curious. the pricing of NAI isn't high enough to make it worth it to rebuild it for anything other than fun, plus we should strive to support cool projects like this
- Mar 1, 2021
- 12
- 10
for NVIDA
You must be registered to see the links
for AMD
You must be registered to see the links
these are both for the text2img and despite what people are saying here I have a fairly outdated PC and have no problems with it.
You don't have permission to view the spoiler content.
Log in or register now.
- Jun 2, 2018
- 1,481
- 5,695
- Jun 2, 2018
- 1,481
- 5,695
For instance, in step 3 (I'm using NVIDIA in my laptop) do I need to download the "official stable diffusion" as well as " Waifu Diffusion Final " or can I just download " Waifu Diffusion Final. As well as in step 5 I can just skip it? Since it is for windows 7 and I'm on windows 10.
I may sound retarded here but I don't want to fuck up the installation lol.
Shit like this seems super interesting and might be good for creating a game here or at least enhancing a game I make.
I may sound retarded here but I don't want to fuck up the installation lol.
Shit like this seems super interesting and might be good for creating a game here or at least enhancing a game I make.
Last edited:
- Mar 1, 2021
- 12
- 10
I can confirm I'm fairly retarded and still managed to set this up
The stable diffusion is basically the same thing as the NovelAi diffusion but it sucks ass, feel free to skip it. I can't speak about the Waifu Diffusion because I haven't tried it. Also Python is needed for all versions, it's just hyperlinking to the Windows 7 version.
It doesn't need an internet connection
Yes
You don't have permission to view the spoiler content.
Log in or register now.
Last edited:
- Jun 2, 2018
- 1,481
- 5,695
Which of these models (as you said you havent tried waifu diffusion yet) did you use to create this booba? Also do you know which one is the best at making hentai and handling sex?
- Mar 1, 2021
- 12
- 10
All the images I've uploaded here are from the leaked NovelAi model. I haven't really been using it for coomer shit but I'd imagine the leak is your best bet for rendering hentai, you'd have to try downloading the Waifu models to see how they compare.
You don't have permission to view the spoiler content.
Log in or register now.