Others [AI] NovelAi and Stable Diffusion being used on games on DLsite

[Darksshades]

Not sure if this is the right place but....

I just noticed that there are now several dlsite games/artist cgs on dlsite that contains the following message

・本作品はNovelAIで制作した画像に対して、効果や効果音、セリフを追加したものを使用しています。
(This work contains images, sound effects, dialogues made by NovelAI)
https://www.dlsite.com/maniax/work/=/product_id/RJ435502.html

or

 本作品はnovelAIにて画像を生成し、加筆修正を行ったものです
(This work contain images made or corrrected by novelAI)
https://www.dlsite.com/maniax/work/=/product_id/RJ429532.html

or even

イラスト:キ印猟品/NovelAI

この作品はAIで生成されています。
(Ilustrator: Kijiru shiryouhin/NovelAI
This work was made with AI)
https://www.dlsite.com/maniax/work/=/product_id/RJ440173.html

And here is a list with all dlsite products that have the disclaimer of having used AI on the description (

You must be registered to see the links

)

Which really surprised me.... I didn't think we had come to a point where AI could be used in a game in this way.
Its not really clear how much of the images/dialogue in those are made using AI but I'm almost sure it meant the CG images, at least in some of the cases.

Checking out novelAI and stable diffusion latest builds. It seems they have now the capabilities to have temporal cohesion and can generate the same "person".

NovelAI is a GPT-based chat and TTS. It has its own internal memory so you can literally just tell what you want and it works, its really like magic... But its paid on a subscription.

But we also have stable diffusion which is free and open source and some of those dlsite games have the artist listed as 'novelaidiffusion' too.
I've seen before that stable diffusion have the capability of saving and generating the same person using DreamBooth which needs to be tested if it works for 2D too.
I have installed stable diffusion here last year and it really does not need a high-end PC, I can generate images with a Nividea 1070 4GB just fine. It takes about 15-20s to generate each image.
I've never generated nsfw images with stable diffusion but its clearly possible seeing as some on those dlsite products, commercial products, are using it.

One big note.
Diffusion AI 2.0 (the hugginface training model) has stated that they will no longer upload NSFW training data.
One can technically still make their own training model but that is not realistic possible for a normal person.
So for NSFW images you'll either have to wait/find some 2.0 third party training model that has nsfw or keep using 1.0.

 

[Tompte]

These developers are exploiting the ability to crank out multiple games in a very short time, hoping people buy enough of them to make a profit, which won't be hard. I predict these types of games will appear on Steam as well. It's just run of the mill shovelware except it just got WAY easier to produce than before.

Pixiv and other image hosting sites has gotten overrun with spam for the same reason. You don't need to be an artist any more to post art, and you can crank them dozens in a single day.

Some developers have been uploading multiple games per month. If you think this is a good thing, you should think a little harder.

 

[Reaper9988]

Late to the Party. But yeah it's gotten annoying, scrolling through blocked images (because I haven't found out how to completely hide AI Generated work only how to block the images) is getting aggravating.
And all of that AI Stuff basically looks bland and the same.
So yeah not a good thing, not at all.

 

[gakgami]

Small note
With these games, an AI generated image is enough to splice up into an AfterEffects/Spine/Live2D animation, so you can be your hunky dory ass the line's gonna get way blurrier than purely still CG.

DlSite now has a little checkmark to say whether we want neural network results shown, and even results that used it but not for everything.

 

[Reaper9988]

Small note
With these games, an AI generated image is enough to splice up into an AfterEffects/Spine/Live2D animation, so you can be your hunky dory ass the line's gonna get way blurrier than purely still CG.

DlSite now has a little checkmark to say whether we want neural network results shown, and even results that used it but not for everything.

Hmm haven't thought about animations actually but you got a point. Can only hope most of the content is tagged correctly, would rather not support AI "Art" in any way.

Also is there a way to actually hide AI Generated Content ?
I blocked the Tag it in the Blocked content list but it still shows the Content, just the picture is blocked.
At least I can see what has the Tag and ignore it(But it's still uselessly fills up the search), I'd rather not see it at all.

 

[Nitwitty]

Not sure if this is the right place but....

I just noticed that there are now several dlsite games/artist cgs on dlsite that contains the following message

・本作品はNovelAIで制作した画像に対して、効果や効果音、セリフを追加したものを使用しています。
(This work contains images, sound effects, dialogues made by NovelAI)
https://www.dlsite.com/maniax/work/=/product_id/RJ435502.html

or

 本作品はnovelAIにて画像を生成し、加筆修正を行ったものです
(This work contain images made or corrrected by novelAI)
https://www.dlsite.com/maniax/work/=/product_id/RJ429532.html

or even

イラスト:キ印猟品/NovelAI

この作品はAIで生成されています。
(Ilustrator: Kijiru shiryouhin/NovelAI
This work was made with AI)
https://www.dlsite.com/maniax/work/=/product_id/RJ440173.html

And here is a list with all dlsite products that have the disclaimer of having used AI on the description (

You must be registered to see the links

)

Which really surprised me.... I didn't think we had come to a point where AI could be used in a game in this way.
Its not really clear how much of the images/dialogue in those are made using AI but I'm almost sure it meant the CG images, at least in some of the cases.

Checking out novelAI and stable diffusion latest builds. It seems they have now the capabilities to have temporal cohesion and can generate the same "person".

NovelAI is a GPT-based chat and TTS. It has its own internal memory so you can literally just tell what you want and it works, its really like magic... But its paid on a subscription.

But we also have stable diffusion which is free and open source and some of those dlsite games have the artist listed as 'novelaidiffusion' too.
I've seen before that stable diffusion have the capability of saving and generating the same person using DreamBooth which needs to be tested if it works for 2D too.
I have installed stable diffusion here last year and it really does not need a high-end PC, I can generate images with a Nividea 1070 4GB just fine. It takes about 15-20s to generate each image.
I've never generated nsfw images with stable diffusion but its clearly possible seeing as some on those dlsite products, commercial products, are using it.

One big note.
Diffusion AI 2.0 (the hugginface training model) has stated that they will no longer upload NSFW training data.
One can technically still make their own training model but that is not realistic possible for a normal person.
So for NSFW images you'll either have to wait/find some 2.0 third party training model that has nsfw or keep using 1.0.

I think Unstable Diffusion is working on the problem. Look them up. I think they are working on a fork of advanced models that won't be NSFW filtered.