I am retarded and I need someone to teach me how to fish. I can't figure the most optimal way to do this.
Suppose that I have a tag list like this, put together with scrapped tags from danbooru, rule34.xxx, sankaku, etc. with maybe a few personal additions:
"[artist name], 3d, (these two are the activation tags and won't be shuffled. I am also adding 3d because I want the model to draw from its existing knowledge of 3d styles), rachel roth, raven \(dc\), raven, 1boy, 1girl, big penis, blowjob, breasts, cheek bulge, completely naked, naked, completely nude, eyelashes, fellatio, female, female focus, fit, fit female, forehead jewel, grey body, grey skin, grey-skinned female, hair, head grab, holding penis, hourglass figure, huge breasts, human, large breasts, big breasts, legs, light skin, light-skinned female, hetero, lips, nude, nude female, nude male, oral, oral sex, pale skin, pale-skinned female, penis, big dick, large penis, purple hair, running makeup, running mascara, runny mascara, runny makeup, ruined makeup, medium hair, sweat, sweaty, thighs, waist, uncensored, bare arms, bare shoulders, eyeliner, eyeshadow, areolae, male, kneeling, looking at viewer, drooling, bodily fluids, saliva, saliva trail, simple background, voluptuous, voluptuous female"

I want this to be trimmed under 75 tokens (currently at 119) for best results with clip (Or perhaps less if you deem any of the remaining tokens unnecessary or counter-productive). But I don't know what to prioritize when trimming. Can anyone take a minute or two and write down how they would simplify it?
I suppose just pruning synonymous tags might be the most straightforward approach, but is it possible that keeping synonyms for "prominent" tags like say "huge breasts, large breasts" while purging less "prominent" tags like "eyelashes" can provide better results for the encoder's understanding? (Keeping in mind that this is going to get compressed 64 times and there will a few pixels left of eyelashes.)