Still very much a WIP but here is a basic template for this model that should work for most styles / prompts. I could fill a page with more tags but they could conflict, it really depends on what your after. Do your own testing, SOME but not all e621, derpi and SDXL tags work.
This is for comfyui btw which weighs () differently just to warn so you might have to play with it for automatic1111 or such. Also if you have too few tags you might need to lower the weight of the quality tags to not get the “overexposed” look. A perfect balance of weights is a constant per prompt fight.
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up, source_pony, ((perfectly drawn)), description of what you want, tag1, tag2, tag3…
Negative:
(((((bad anatomy, bad proportions, bad quality, glitch, blurry))))), watercolor painting, brush strokes, signature, watermark, borders, abstract,
You might also have better luck with fewer “score” tags such as only score_9, or score 9,8 and 7 but not the others. It depends on how “out there” the prompt is that you might need a greater width of scores to draw from for some rare concepts not tied as well to better scores.