The 22 Billion parameter problem:

• If I look at one image per second every waking second it would take 697.6 Years to look at a 22 billion images.

• The answer we "agreed" upon was to have AI assign ethnicity, age, gender etc.

• The outcome 60%-80% accuracy on the largest LLM's

Will you post the results?

SD 1.5, SDXL, FLUX, ask anyone of these models for an image of a "nude, flat chest, female" Run a batch of 100 images. Will you post the results?

Why does flat chest cause more issues? AI doesn't care if it is a boy with a flat chest, a drawing of a girl with a flat chest, or perhaps on of the countless images of children that was miss-marked or just under tagged.

• Every human is effected by anchor fallacy, I asked for x, there for the image is x

• Moderator's can be effected by it if they know an image is reported. It was reported therefore it is bad.

I do not have a solution for the problems above.

I have commented on appending 1girl to female or woman (For FLUX anyway) but this doesn't address the greater problem that comes from a high parameter model.

I have read the comments about 1000's of images getting removed and I agree with the frustration. As to me this appears as one of two ways.

1. Civitai was incompetent and let CSAM images on the site for months

2. Civitai is putting up a show for the "Open Model Initiative"

As a creator I also find the TOS section (C) disturbing

"(c) your User Content could not be deemed by a reasonable person to be objectionable, profane, indecent, harassing, threatening, embarrassing, hateful, or otherwise inappropriate."

I am part of the top 2% user engagement score creators now, but who's to say that tomorrow my content won't be found objectionable.