Type | |
Stats | 356 323 |
Reviews | (56) |
Published | Sep 5, 2024 |
Base Model | |
Usage Tips | Clip Skip: 2 Strength: 1 |
Trigger Words | 0nsenl4dy dark brown eyes, big eyes large breasts |
Training Images | Download |
Hash | AutoV2 03D813CBF2 |
Pony v2 Update (05/09/24):
better tagging and dataset this time
strength: 1
breast size can be either 'large breasts' or 'huge breasts', kiwisoda's original images seem to be inbetween.
trigger word is: 0nsenl4dy
additional words: dark brown eyes, big eyes
adetailer is recommended however if you want the eyes to have a more watery look, leave adetailer off
In the main showcase I've used a bunch of realistic pony models to generate and in the gallery below there will also be anime model generations to see how it looks.
Pony SDXL update:
This is an experiment to see how well i can train pony xl
you can leave it as it is for a slightly illustrated/anime look, or use 'realistic, photorealistic' for a closer approach to the original
I wanted to train on an anime checkpoint so I don't have to keep switching parameters between models
trigger: PD3-OL
weight: 1.0
the trigger word should cover the basic features but you can also include 'dark brown hair, brown eyes' if it doesn't show up
Pony edit1: Thanks to 'XXXNOAHXXX' you can also use the following to achieve a realistic look:
((realistic photography:1.4, photo realism:1.4, realistic detailed skin, realistic detailed eyes, sharp focus, high resolution, masterpiece, highest quality, absurdness, 16k))
\\\\\\
SD1.5 v1:
A lora inspired by two images generated by kiwisoda1010:
links to full images (nsfw):
Image posted by kiwisoda1010 (civitai.com)
Image posted by kiwisoda1010 (civitai.com)
\\\
strength: 0.9-1.0
trigger: brown eyes, bangs, large breasts
depending on the model, may need to add: (realistic, photorealistc)
'bangs' is optional, can be replaced with (but not limited to):
forehead
parted bangs
crossed bangs
swept bangs
asymmetrical bangs
adetailer was used in genning.
Used the same model and similar settings to gen dataset.
Note: the data set was sorted out manually and is prone to human error, so the final result probably won't be able to perfectly replicate kiwisoda's images.