Experimental LoRa, in the sense that I had no idea what the outcome would be.
The images used have a very specific style (found by chance in 2024 while running tests), and for the trigger words, I used the prompts that were used for image generation without knowing if the AI would actually understand them.
And the initial results seem quite promising.
Some can be seen here:
https://pixai.art/fr/@user-1700304173167566330/collection/1736971801481979695#before=gaJpZNcAGleFcA%2FFK3A%3D
Title: Fantasy.
Trigger words: Chatfant, (fantasy castle), (fantasy medieval painting style), (fantasy medieval illumination), golden, brillant, ultra detailed, intricate details
"Frame" can be used in both prompts and negative prompts.
Negative prompts (with some models): (high view:.9), (aerial view:1.9)
The risk of aerial view is due to the model used and not to the LoRa.
Strenght: 0.7 -> 1
