"aesthetic 11" tag
I saw somewhere that people use "aesthetic 11" as a style tag and there is even a link to someone claiming them to be official tags during training.
I tried it on my images and it made the images more vibrant but lost a lot of prompt adherence and caused mayor concept bleeding:
Plantfolk on the left and catfolk on the right...
Without the tag:
plantfolk on the left and human on the right (still a bit of bleeding going on with giving the human pointy ears) but they are more easily distinguished .
Is this tag an official thing or just someone mistaking models and tags?
it's official tag
there's aesthetic rank from 0 to 11
where 11 is geared towards curated synthetic data
10-0 is real data but for aesthetic 10 there's some mutation chance that it will mutates into aesthetic 11 tags
Tested the aesthetic from 11 to 7 and while it makes colors more vibrant and adds contrasts, the adherence to the prompt suffers. Especially concerning multiple subjects in the scene with different descriptions like skin color, or fantasy features like fur or feathers, those disappear or get used on all characters when the tags are used.
Without the tags, there is a bit of concept bleeding but not to the same extent.
So in conclusion, for my prompts and setup, the "aesthetic" tag is not so good as it does something that lets the prompt adherence suffer on multiple subjects. It help on single subject prompts in the described ways (better vibrance and contrast). Something to ponder and use for specific prompt situations, sadly for me not a thing for every prompt...
Thank you so much, this is really useful information, have you tried adding low "aesthetic" tags to negative promptings, can this improve generation?
Thank you so much, this is really useful information, have you tried adding low "aesthetic" tags to negative promptings, can this improve generation?
Absolutely, I use 7>1 as a default, it helps shift it away from the fluxed Midjourney-like results with overly warm colours and smooth look. then you should expirement with other additions for whatever generally helps with your prompts. "aesthetic 10, photo_medium, a woman standing in the streets of london crowded with people, she is jogging in a hurry, the sky is blue and the sun is shining, aesthetic 10, photo_medium"
But that said, negatives don't work as we want them to work, take a look here how Even though #2 and #4 are compositionally very similar they are completely different negatives, and how #1 and #3 are nearly the same even though it's just #2 with even more negatives to it.
Just like with SD1.5 and SDXL negatives just rummage around different interpretations that you could also achieve with slight changes to resolution, CFG, steps, clip skip, etc.
