This model changes the level of detail
Detail tweakers aren't new, but here we go anyway. This model increases the number of intricate lines and the complexity of shading, essentially it is a realism LoRA, the higher the weight it is at, the more features will be present in the image with marginal impact on the composition.
Inspiration
I just noticed there wasn't a detail tweaker for Pony Diffusion XL yet and I decided it was prime time to train one. it was quite easy to train and only took about an hour because this is the first and only version I made, if there is an abundance of issues that I haven't recognized thus far, I will make updates, but as it stands it seems pretty good I think.
Donations
Speaking of training, training models is expensive and I run training on my own private server, if you like what I do, consider supporting the development here!
Key Benefits
allows the alteration of details on a finely controlled level
far more predictable than just prompting for "detailed" and praying to a deity of your choice
allows for more extreme detail than would be possible with mere prompt weighting
Quirks
I would recommend putting "realistic" and "3d" in the negatives for V1 because V1 has a tendency to make some uncanny valley things at higher weights.
V2 seems to function exactly how I would want it to. if you encounter any strange behavior please let me know.
Description
Trained for 1000 steps.
used a now outdated training script for my LECOs, I might revisit it later if there is a demand for it
FAQ
Comments (8)
this is wonderful!
im currently testing on my own, excited to see the outcome
you might tempt me into a friendly competition to see who can make the most robust detail tweaker lol, there are problems with this one, but I have learned how to fix literally all of the issues in the training since then, so I might revisit it if people thing the issues need fixing (I don't think the issues are very severe)
can't wait to see what you've cooked up though, I'll keep an eye out!
@Yolup can't wait to see a new detailer if u make one, cause i love this a lot.
how does one train a detailer lora? is it specific to data images ? or is there more? asking cause i wanna try making one too
I used the LECO training script on GitHub by P1atdev based on the LECO paper to train this model, the LECO training process generates an image at an arbitrary denoising strength, and then trains on the difference between the output of the model when prompted for a concept, and when unprompted. this allows the model to alias tags, words, concepts, or phrases to any arbitrary prompt, in this case, I aliased
"detailed"
to
""
so I effectively was training it to always produce images that looked as though it had been prompted for "detailed"
I did some other cool things myself, but I can't fit all my information in here, I should make a tutorial sometime though.
@Yolup ok thanks ya a tutorial would help me understand it better. can u post link to Leco training script if there is a link?
@Darkhitman https://github.com/p1atdev/LECO
@Yolup thanks
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.