- Dec 31, 2016
- 316
- 708
Combining loras or similar things, often causes conflicts like oversaturated colors, disfiguring etc mostly cause they are poorly trained or you are using ones that are specifically made to target the same thing.
Generally you should be pretty safe to use lora/TI/etc at 1 or close to it on it's own, the further away from 1 you need to apply it the more poorly trained it is (not the case for more general styling though where weight is more how "hard" you want it to apply, IE dim lighting).
If you have a lora that you need to drop to as low as 0.2 you more than likely have a perfect example of something that should just be deleted and trained again.
You can use weighting to figure out how to adjust you train as well, think of it as value to apply to your learning rates (or epoch/repetitions), for lycoris this gets even more accurate as you can adjust weight for both values and make it easier to adjust both learning rates.
And now, something completely different...
Generally you should be pretty safe to use lora/TI/etc at 1 or close to it on it's own, the further away from 1 you need to apply it the more poorly trained it is (not the case for more general styling though where weight is more how "hard" you want it to apply, IE dim lighting).
If you have a lora that you need to drop to as low as 0.2 you more than likely have a perfect example of something that should just be deleted and trained again.
You can use weighting to figure out how to adjust you train as well, think of it as value to apply to your learning rates (or epoch/repetitions), for lycoris this gets even more accurate as you can adjust weight for both values and make it easier to adjust both learning rates.
And now, something completely different...