t.me/oaf40_pub
Description
Model used: Rainpony XL Rainfall (v2)
My telegram channel: @oaf40_pub
In this project I've been experimenting with achieving decent dark backgrounds. Considering my project creation work flow it was an interesting challenge. No "darkening" or "night light" LoRA has been used to create this image this time. The part of the prompt responsible for the initial background:
[dark background, black background:outside, rooftops, city, sky, night, stars:0.2]
Such a contraption switches from using dark background, black background to outside, rooftops, city, sky, night, stars at 20% (0.2) of inference progress, i.e. if you have 40 inference steps then the switch will happen at 8th step. This allows bringing extra emphasis at various aspects of an image.
However at the high-res stage of my project creation flow the checkpoint&style LoRA tandem kept bringing the contrast back to its boring bright look even with aforementioned prompt trick applied. I found no better way to defeat it than letting the model do its nasty thing then lowering the overall exposure and shifting the Black Level in GIMP using the "Exposure" filter on the last step of working on this project.
For reference, here's my previous project where I've been working on the dark background using the "Low Light" LoRA and some LUTs applied in Krita: Going to sleep when it's raining outside (e6ai post)
There are no visible comments.
Login to respond ยป