>>11338998
TOP KEK that anon finally got his insane request done, 11/10 gen
>>11339069
hahaha nice, glad you liked them!
>>11339109
>style mining?
Not much to add to blurry eyes anons reply other than I also like to browse danbooru for tags and related artists with this page: https://danbooru.donmai.us/related_tag?commit=Search&search%5Bcategory%5D=Artist&search%5Border%5D=Frequency&search%5Bquery%5D=inverted_nipples+
In general you can also try to eliminate as many quality tags and negs as possible in your prompt, that also helps to get away from a generic look. Also some models and artist tags allow you to go pretty high on the weight. Don't shy away from going crazy, I sometimes use specific tags at (:1.4) or even (:1.6) lol
>>11339764
>too big
Hard disagree, it's perfect.
>>11339972
very cute
>>11338022
I couldn't help myself and got easy lora scripts installed, so I can give it a go to train some v-pred loras. But I ran into a few issues lol Biggest one was that I needed a newer python version that was unsupported by the autotagger UI that I liked to use. I made do with some janky comfyui workflow but I wanted to ask how you tag your images? So far it worked for me but it's not ideal since I can only use one tagging model. HWtagger that I used would use a few models and prune the less releavant tags automatically. That said, I'm kinda suprised how well things worked for me. I got a few styles baked already.

If anyone likes I'd love to hear your feedback on this hayashibara hikari style. But fyi, it's a heavy loli style and also merged so the baseline strength is still 0.4. Picrel used that lora. You can find it in my rentry.
>Temp share folder
>hayashibara_hikari_noob_vpred_test_merged
>uma girls horse cock cumming
https://rentry.org/0point4_anons_AI_rentry
>>11339698
lol maybe picrel is close enough. I've put a few more tests in that share folder mentioned above.