READ ME, READ ME, READ ME!
To avoid an over-baked LoRA look, unlink Unet and TE with the syntax <lora:LORA:X:Y> where “X” is the Unet weighting and “Y” is the TE weighting. For this LoRA you generally want a your Unet to be higher than TE, see example prompts below. Aspect ratio is 512 x 768.
This LoRA is meant to add a pendulum / coin on a string hypnosis theme into your generations. Due to the limitations of SD 1.5 it is still a little RNG dependent, but overall I have found it more consistent and higher quality than other similar LoRAs.
General safe weighting: <1.0:0.8>
– <0.8:0.6>
. Lower weighting will effect the original image less but will have more RNG whether it produces the pendulum / coin on a string.
This LoRA can also do the “?”, “??”, “blank stare”, “cross-eyed”, “hypnosis”, and “mind control” tags pretty decently as well if you are using a quality checkpoint. I think this LoRA turned out pretty well given the rather limited training dataset (~40 images).
Full example prompt: StSPendulum, (swinging pendulum, coin on string), mature female, ?, ??, ((head tilt), blank stare), (cross-eyed), hypnosis, mind control, gym uniform, open mouth, [drool], [ahegao], arms at sides, facing viewer, blue hair, green eyes, medium breasts, locker room, highres, absurdres, masterpiece, best quality, ultra-detailed, illustration,colorful, masterpiece, best quality, ray tracing, 4k, (extremely detailed), soft lighting, tracing, soft shadows, detailed face, <lora:StS-Hypno-Pendulum-Coin-Iter5:0.6:1.0>, <lora:more_details:0.4>, <lora:add_detail:0.4>
(“StSPendulum” is my activation tag. It is not necessarily needed).
KNOWN ISSUES:
The string of the pendulum tends to blend into background elements
There may be a phantom string of the pendulum generated
Edges of the pendulum’s coin may be a bit blurry
LoRA too easily yields “overbaked” look.
Changelog:
Feb. 21, 2024 – Inital LoRA release