Type | |
Stats | 592 1,178 |
Reviews | (10) |
Published | Jun 26, 2023 |
Base Model | |
Training | Steps: 3,360 Epochs: 10 |
Trigger Words | xsv5_style |
Hash | AutoV2 E438C79AA4 |
"It's about the journey, not the destination."
Heya, if you like my work, feel free to buy me a Ko-Fi! It is greatly appreciated, and it makes more of my work possible! <3
The first of the NijiLike series, NijiLike V5 LoRA is based on my own generations with Nijijourney. It is not related to any of the other Nijijourney/Midjourney generations I've seen out there, but the more Nijijourney styles we have, the better I suppose! NijiLike works best on anime models, this is tested well with MeinaMix!
This version of NijiLike is trained with 28 images, providing a direct, focused anime-like output, inspired by modern anime screencaps. The trigger word is xsv5_style, though it can probably work without it? For more controlled outputs, I recommend setting the weight to 0.4 to 0.6, though feel free to experiment to your liking!
Storytime:
It is my first LoRA to be developed, thanks to holostrawberry's LoRA generation tutorial, which helped me a lot here. As a side effect, while I did implement a global_activation_tag to it, it's mostly optional and it's just me trying to get around things. Sometimes people are bound to make mistakes.
I also do not have a good PC to develop faster images, so these things took quite a lot of time (around 1 - 4 minutes per image, sometimes even more) to even generate one image of this. But regardless, it is a community effort after all.
Hopefully, all of these images that I've generated will be put into good use. And there will be more NijiLike LoRA to come!