Type | |
Stats | 558 7,938 |
Reviews | (45) |
Published | May 27, 2024 |
Base Model | |
Usage Tips | Strength: 1 |
Hash | AutoV2 88172C0F5D |
Note: the title has been changed: Abyssorange Style -> Pony Style 2.5D
I realized that training using SD1.5 generated images will inevitably mess the details, after playing with the training configs.
The new version is using semi-realistic illustrations to approach. As a result, it may not faithfully reproduce the AbyssOrange style.
Therefore I have changed the title and made the previous LoRA a sub-version.
Version <25d_528>
This version is providing a 2.5D style on AutismMix and it makes better details comparing to the previous version.
No triggers needed.
You can make combined use of the two versions, for example: <lora:style_blend_25d_528_d32:1> <lora:abyssorange_style_for_pony_d32:0.5>
Verison <abyssorange_approach>
This is not an official AbyssOrange XL model!
This is a trigger-free style LoCon (technically, LoRA with full layers) that works for PonyXL / AutismMix, following the style of AbyssOrangeMix2.
No triggers needed, but more precisely it's bound to quality words: score_9_up score_8_up score_7_up. source_anime / negative realistic is optional.
It's trained on generated images using AOM2, mixed with some other aesthetic LoRAs and LECO fixes.
Since it's trained on SD1.5-based AI generated images, it can negatively impact generation quality on details, such as blurry or broken fingers! Be careful when you make merged models upon this LoRA.