I'm annoyed by NovelAI. Currently, the limbs are drawn strangely, but it will be difficult to distinguish them in the future. We also need to set the "novelai" tag.
もちろん、管理者がAIで生成した画像のアップロードを許可しているかどうかにもよります。 Danbooruからは今のところそのようなtagは見当たりません。saemonnokami said:
I'm annoyed by NovelAI. Currently, the limbs are drawn strangely, but it will be difficult to distinguish them in the future. We also need to set the "novelai" tag.
近い将来には高解像度の画像も作成可能になるかもしれない。あるいは、AIの画像を加工して、複数のレイヤーとして使用する場合もあるかもしれない。NovelAIの画像のアップロードを承認するのか否か、あるいはFaultsタグを設定するかどうか、コンセンサスが必要だと思います。whitespace1 said:
もちろん、管理者がAIで生成した画像のアップロードを許可しているかどうかにもよります。 Danbooruからは今のところそのようなtagは見当たりません。
近い将来、人工知能は労働集約的な仕事に徐々に取って代わり、その背後にある商業的利益は、投資家にとってより重要なものとなるでしょう。時代が進化し、コードが最適化され、量子コンピュータが開発される中、次世代にどれだけ心のこもった本物のアートを見ることができるのか。人間は時間とともに老化し、一生のうちにできることは限られていますが、機械や人工知能だけは進化を続けます。
Stable Diffusion has great potential but Novel AI's training set isn't quite there yet.
It mainly uses tags from Danbooru etc. as labels which are far from perfect. They're not exhaustive, errors are everywhere, and plenty of tags were "misunderstood" due to how gereneric they are.
It also has funny side effect that some seemingly inconsequential tags will dramatically change how the results look like and narrow down the result artstyle to a very limited pool. For example, it generally can understand hair color, outfit type and color relatively well. But using only generic NSFW tags (say, sex) as prompts often cannot generate proper images due to how these tags are used on too many posts for the AI to find anything coherent to recreate. You have to constantly throw super specific (which you don't even want) tags to have a good result, but then they all look kinda the same...
I think it at least needs 1) a larger dataset 2) a good natural language preprocessing on the tags/labels.
It mainly uses tags from Danbooru etc. as labels which are far from perfect. They're not exhaustive, errors are everywhere, and plenty of tags were "misunderstood" due to how gereneric they are.
It also has funny side effect that some seemingly inconsequential tags will dramatically change how the results look like and narrow down the result artstyle to a very limited pool. For example, it generally can understand hair color, outfit type and color relatively well. But using only generic NSFW tags (say, sex) as prompts often cannot generate proper images due to how these tags are used on too many posts for the AI to find anything coherent to recreate. You have to constantly throw super specific (which you don't even want) tags to have a good result, but then they all look kinda the same...
I think it at least needs 1) a larger dataset 2) a good natural language preprocessing on the tags/labels.
AI Art is quite the can of worms. Danbooru has a hard ban on it, with the exception of "AI Assisted." There is now a separate board for AI artwork, https://aibooru.online/posts
I am opposed to the images produced by AI appearing on the board, these images are good to keep for yourself.Genex said:
AI Art is quite the can of worms. Danbooru has a hard ban on it, with the exception of "AI Assisted." There is now a separate board for AI artwork, https://aibooru.online/posts
有个AIbooru也挺好的，免得给yande.re引来太多的版权麻烦。我比较反对在yande.re上传AI作画。Genex said:
AI Art is quite the can of worms. Danbooru has a hard ban on it, with the exception of "AI Assisted." There is now a separate board for AI artwork, https://aibooru.online/posts
That's by design, NovelAI trained this model so that it can believably-ish generate the same character in different situations, since their AI art is intended as illustrations for generated stories.fireattack said:
and narrow down the result artstyle to a very limited pool
wahh
This artist is "AI illustrator".
https://yande.re/post?tags=mjonir
Should we add tag like "AI Generated" to identify AI arts?
https://yande.re/post?tags=mjonir
Should we add tag like "AI Generated" to identify AI arts?
Nope, as they're fake artworks not drawn by human beings.Madao said:
This artist is "AI illustrator".
https://yande.re/post?tags=mjonir
Should we add tag like "AI Generated" to identify AI arts?
whitespace1
NovelAI
https://forum.gamer.com.tw/C.php?page=3&bsn=60076&snA=7378038
https://cyberes.github.io/stable-diffusion-models/
AI繪圖開始在Pixiv大量產生，如果作品不打上NovelAI、NovelAIDiffusion標籤很難一時間辨別出到底是手繪的還是機器畫的
https://www.pixiv.net/users/86933000
https://www.pixiv.net/users/3750014
https://www.pixiv.net/users/1137830
https://www.pixiv.net/users/86889740
重點是輸出的關鍵字及各種調校，還有一台不算太差的電腦配備(主要還是N卡
我還是看看就好，年紀大了，想做什麼都覺得好懶...