This is my current point of frustration with SD and I'm wondering if it's an intractable problem or I just haven't figured out how to engineer the prompt to do this yet. Getting two different characters in the same image each of which is a different species, with no bleedover of traits between one and the other. The only way I've figured how to reliably do it so far is with regional prompting in comfy, which is fine so long as the characters don't overlap nor interact with each other at all. If you actually want them doing stuff together then it all seems to break down and you get catdogs and dogcats and usually their limbs are fused together and etc. I've seen images with A1111-style prompts with BREAK, END, ADDCOL, etc, trying the prompts myself doesn't seem to result in anything more reliable than without that syntax in it. Neither do bracketing, line breaks, or anything else seem to help with chunking SD's attention the way I want to chunk it.
Long shot but do any AI artists have tips for this sort of thing? Browsing the ai_gen tag it does seem like 90% of images are either a) solo shots or b) same-species multi character compositions. Very rare to see two separate species rendered correctly in the same image. I've noticed even in the stuff I made so far, when there are two or more characters of different species it's usually just a) an adult that absorbs all the character attributes in the prompt, and then b) a generic nonspecific flesh-toned cub, like barely nonhuman even.
| Viewed: |
36 times |
| Added: |
8 months, 1 week ago
04 Sep 2023 09:12 CEST
|
|
|
