this textual inversion embedding was made with SD 1.5
has more zazz with Protogen x3.4 (as seen with the example images), probably works nicely with other custom models too
dataset was mostly landscapes, does influence other things tho, but might have a weaker effect on them
not great with quokkas
place evreka15.pt or evreka15.png in your embeddings folder
use as “art by evreka15, [your prompt]"
it does work better with that structure as opposed to just using "evreka15". this is due to the filewords training afaik, where the prompts in the template follow that syntax
if the style doesn’t come through strong enough it helps to add “painting of”
, so the prompt would be “art by evreka15, painting of [the rest of your prompt]”
or “painting of [your prompt], art by evreka15,”
(these both express the style differently, most of my examples use the former but you might enjoy the latter more)
negative prompts that can help: border, frame, blurry, pixelated, low quality
all examples use Euler_a
steps between 20 - 52
CFG scale 7 - 10
made with highres fix in the A1111 web ui
✦•······················•✦•······················•✦
if you enjoy this consider buying me a coffee (ノ◕ヮ◕)ノ*:・゚✧
✦•······················•✦•······················•✦
CreativeML OpenRAIL-M license applies, the most important points are:
You can't use the embedding to deliberately produce nor share illegal or harmful outputs or content
The authors claims no rights on the outputs you generate, you are free to use them and are accountable for their use which must not go against the provisions set in the license