Neural network

[Yamych]

Can anyone recommend a neural network for generating hentai art?
I use Seeart with a paid subscription, but to be honest, I don't really like it because the art is too monotonous.
If there is a better neural network, how can I transfer my previous characters to it?

 

[osanaiko]

This fairly recent thread is a good place to learn: https://f95zone.to/threads/your-noob-to-expert-guide-in-making-great-ai-art.256631/

This is the OG thread where several users shared their learning process when the Stable Diffusion tools first arrived: https://f95zone.to/threads/stable-diffusion-prompt-sharing-and-learning-thread.146036/

 

[akeira]

Hi

This fairly recent thread is a good place to learn: https://f95zone.to/threads/your-noob-to-expert-guide-in-making-great-ai-art.256631/

This is the OG thread where several users shared their learning process when the Stable Diffusion tools first arrived: https://f95zone.to/threads/stable-diffusion-prompt-sharing-and-learning-thread.146036/

Hi could I ask the opposite question, is there a thread about AI tools for categorizing/analyzing photos? Like grouping photos/art by type, that sort of thing.

 

[osanaiko]

Yes, but also no. It depends on what you mean by "AI tools for categorizing/analyzing photos? Like grouping photos/art by type"

There are the (now quite old) "CLIP" neural networks, which outputs words and phrases associated with features from pixel array inputs. Those ones works quite broadly but does not have much new training data, and are not specialised towards at "adult themes" - if i remember correctly, it used a bunch of human tagged data as a source so was pretty good for its time. Even today those things are used somewhat in the text-to-image model pipelines during training.

There is also the "danbooru" model which leveraged the hand-annotated data set from the Danbooru website which has 100000s of anime images and extensive tagging. this does reasonably well at the stuff you see in anime style artwork, including artist identification, and is quite good at adult stuff.

Beyond that, I'm not certain what recent **image** classifiers have come out. Many of the multi-mode LLMs now include training on various types of media other than just text, although it depends on the model and company as to whether that is direct pixel-stream training or some sort of conversion-to-tokens processing. Those types of model are generally NOT going to give good results for adult stuff or comic/anime/hand drawn stuff specifically, because they are products from commercial companies who generally avoid adult materials.

What sort of use-case are you trying to solve?