Type | |
Stats | 2,988 55,777 |
Reviews | |
Published | May 19, 2024 |
Base Model | |
Training | Steps: 2,642 Epochs: 18 |
Usage Tips | Clip Skip: 2 |
Hash | AutoV2 48C410236E |
PONY v2.0 UPDATE!
Improved overall quality and more vivid colors.
Try with my Checkpoints Smooth Mix for some nice results.
Use weights around 0.5~0.7 or the colors will go insane! x_x
Both Pony Lora versions look great so just use the one you like best.
Have fun!
FLUX UPDATE!
This one was a pain to make. After several attempts (and over 15k buzz X_X... bruh) I ended up with this version that doesn't use tags.
According to civitai's own guide captionless trainings should work and I went for it. I don't usually like doing this but I guess this work well for Flux.
I would appreciate feedback since this method costs less buzz and is way faster.
Describing the artstyle you want will make all the difference. Using tags like "anime artwork" and "fantasy art" helps alot.
Use 20~30 steps and weight 1~1.2 for creation.
All images were created using flux1-dev-bnb-nf4-v2.safetensors
Have fun!
===============================================
Style-LoRA of the artist/goddess/miracle of the universe Sakimi Chan. If you like this style go check her patreon and consider giving support!
This model works really well with 1280x980 or 980x1280 resolution. Try increasing the resolution with Highres. fix like this!
This one is tricky. Around 0.8 weight the AI loses its mind so the hands start to get messed up. So I suggest using the LoRA around 0.7~0.8 weight for good results. Negatives and combinations with other styles might mitigate that problem.
Enjoy! :)
I use these prompts on every image with great results:
POSITIVE:
score_9, score_8_up, score_7_up,
beautiful aesthetic, very intricate, high quality details,
You can get really good results with no Negatives. Try it out.