This model has been trained on screenshots and artwork from TES Morrowind. Works well with Rev Animated with 1.0 strength. It understands tags naming main settlements and can mimic their style.
Useful tags: balmora, caldera, vivec city, city, village.
Description
This version has milder results, making it compatible with more models (especially realistic ones), but it can also generate less morrowind-looking results.
FAQ
Comments (5)
I keep getting severely burned results! The only way I can get anything useful is extremely low CFG, no other loras, no img2img - I was really hoping to use this with my img2img workflow, is there anything I can do?
I haven't tested any more complicated workflows, I'll have a look at that and see if I can tune it to get better outputs
@jekk That would be wonderful, thank you! I've been blocking out backgrounds in color, then running them through img2img with arty loras to make quick and dirty backgrounds for illustrations, being able to make Morrowind locations would just be amazing.
I've re-run the training with lower learning rates, cleaned up dataset and more steps, can you try it out?
https://drive.filen.io/d/afc73093-28fa-486e-89b2-7d5b4ba42097#9JJYDZW2cWSMKo3U1DpRnJgUUadI8SAz
I recommend mentioning city name in the prompt like 'balmora', 'caldera' or 'vivec city'.
It seems to be playing more nicely now with other loras, though on many models it still burns real easily and really doesn't like having a CFG above 7 at all. There's still some funkyness with my img2img workflow (I block out color zones, then img2img it to get the background) but it's pretty signifcantly reduced.