Wanted to try to reverse engineer the Detail Tweaker LoRA as just a fun experiment and what I got is a lora that raises details and dynamic range while also adding a touch of realism to even none realistic centered models and to my surprise it is also able to do the opposite when lowering the strength removing details and dynamic range and removing the realistic style.
Lora strength can range from -2 to +3, anything above 3 will depend on the prompt and\or the model
Description
FAQ
Comments (13)
Does this make anthro's look more 'feral'?
Not to my knowledge. I believe it is intended to give more detail to fur textures (or just texture detail in general).
Excellent work, great Lora.
summarizing:
Use negative values to get a cartoon or painting style
Positive values, make more detailed, realistic or more feral style
This is an amazing share, do we still need to use the negative embeddings (bwu dfc ubbp updn, duo)
I always do, but thats because i'm a bit lazy with prompting and they seem to work well in most cases
Pls list those negatives
What exactly are (bwu dfc ubbp updn), anyway? I see them used everywhere, but nobody has a clue what they're actually exactly for, and even searching Google does nothing.
[And I would usually not negative (duo), though I understand why you might. The occasional random second characters tend to be quite interesting... even better than intentional second characters... as long as the rest of the negatives are keeping things under control.]
@zareltgr2 They are my negative embeddings you added them to your negative Prompt to help guide the model toward more realistic style and if wondering what they stand for; bwu- stands for blurry, watermark, unrealistic, dfc- stands for dull, flat, color, ubbp- stands for unbelievably bad body parts, updn- stands for ultra-saturated, painting, drawing, not realistic. duo isn't one, but sometimes when making higher res images it would double up the character.
So going negative will make a tiny bit anthro correct? I like em monster girl style
going negative will make it more a simple drawing style and less photo real/sharp
I'm amazed this hasn't been updated to sdxl or pony
I also support the call! Can you find it in your heart to update this for XL models? This is/was an amazing Lora for 1.5.
Can you please share how did you train it?
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.