Update!:

Anima version is released. Since Illustrious was using CLIP (which is ancient technology at this point), Illustrious was pretty bad for this type of concept. It was unable to understand numbers let alone something ambitious like a "virtual dick".

As you know, Anima is using Qwen as the text encoder. A proper LLM with intelligence. It opened the doors (just a little) to this concept. Important: the tag is "erc score=" in this version.

Example: erc score=10,......

This lora is truly a black box. It can change anything to make the image more "goonable". But the core features of the original image mostly stay the same.

OLD (Illustrious):

I made a new stuff, I call it "ERC System".

To explain what is that, I'm gonna ask one question: What makes us think something is lewd?

The answer is simple. We humans have genitalia. And our organs react to lewd things.

AI doesn't have that. Yes, AI knows what is explicit or what is safe, but it doesn't know which porn is better and which porn is worse, it doesn't have feelings. This is the core problem of AI. Big corporations don't want AI to become human-like. They want AI to stay robotic as they are. So I came up with the idea: "implement a virtual dick to checkpoint." This lora tries to serve that purpose.

Similar to rating tags (explicit, suggestive, safe...), there are levels in this lora. From "erc level 1" to "erc level 10". Each image is given a lewd level based on my judgement. You don't have to limit yourself to 10. Further levels are possible like "erc level 80" or "erc level 8775875". But the last level was 10 in the dataset.