extension_region-p.md

I have had enough with SD confusing my prompts and interchanging attributes between objects and subjects so after a short look, i found out the Regional Prompter extension (the extension is available to install directly through automatic1111 or here https://github.com/hako-mikan/sd-webui-regional-prompter) after playing with it for a bit and was glad with the results, i tried to push it further by combining two different concepts (light above water, dark underwater) in the same prompt. this is something that Midjourney failed to do, Dall-e/Bing (which i found to be the most capable in understanding complex promots) was close, but still suffered by washing everything in the same lighting and color and SD is no where near capable doing that based on every attempt i tried. maybe someone could achieve it with clever prompting, but i never managed to do so without the extention.

You can see in the second image the regions settings i had done to seperate the concepts. the regions tend to blend with each other, which can be good if you don't want a very sharp divide between the regions, but it can also affect your results, so i had inserted a few buffer regions to better seperate the two concepts.

Prompt

side view of a giant boulder <lora:sxzBlizzardStyleWarcraft_sxzBlizzV2:0.25>  <lora:mermaidsLoha_v120:1> (pascal campion:0.3) long shot, (side view), lake, masterpiece, high quality  ADDBASE blue sky, bright day light ADDROW side view, above water, lake, bright, clear skies, day light ADDCOL low angle, long shot, yellow clear bright day light, above water, teal lake water,  side view of a (woman mermaid:1.5) with fish tail sitting on a rock boulder ADDCOL lake, above water, bright, clear skies

ADDROW (semi translucent water ripples), foam, transition between above water and (underwater), side view of boulder in the center

ADDROW submerged, underwater, dark ADDCOL long shot, ((underwater)), submerged, deep, dark, side view (glow:0.4), volumetric fog, monolith boulder made from a piles of small bones and many human skulls ADDCOL submerged, underwater, dark ADDROW underwater, sand, bedrock, blue fog, volumetric

Negative prompt

easynegative, nsfw, perspective, ADDCOMM

Settings

Steps: 25, Sampler: Euler a, CFG scale: 7, Seed: 2768402191, Size: 512x768, Model hash: f57b21e57b, Model: revAnimated_v121, Clip skip: 2,

Regional Prompter settings

RP Active: True, RP Divide mode: Horizontal, RP Calc Mode: Attention, RP Ratios: "1;2,1,2,1;1;5,1,4,1;1", RP Base Ratios: 0.2, RP Use Base: True, RP Use Common: False, RP Use Ncommon: True

If you are trying to reproduce the exact image, due note that it fails to generate the skulls at the base of the boulder, but a single inpaint with the BoneyardAI LORA (https://civitai.com/models/48356/boneyardai) at a medium strength did the trick.