Style LoRA of artist MARE.
Description
Lycoris + Smaller
FAQ
Comments (4)
Could you possibly upload the training data for this?
Hi @MidFR. I've been testing your Lora and I see that it has a lot of potential, but I'm going to say something out of place: the V1 version of Pony is better than V2. This happens because V1 seems more trained than V2. I can leave some suggestions for V3:
- Start from V1 in terms of training.
- Allow to reduce the strength of the Lora (currently it fails if you choose 0.8 or 0.7 strength, for example). This is very important to make the style Lora flexible with other Loras.
- Increase the amount of trained tags.
V2 was trained with more steps and trained as a LyCORiS instead of a LoRA (it just doesn't work on civit generator if I set it as a LyCORiS), the training data was the same in both including tags. This style is one of the styles when I tested that really needs to run "3d" in the negatives or it will not work.
I see you're training Flux LoRA's now (Milada Moore) and I just wanted to say this is one of my favorite ever for Pony and it would make a fantastic addition to Flux. Thanks again for this!
Details
Files
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.