Type | |
Stats | 230 389 |
Reviews | (47) |
Published | Jun 12, 2024 |
Base Model | |
Training | Steps: 3,200 Epochs: 10 |
Usage Tips | Clip Skip: 2 |
Trigger Words | Cybersigilism |
Hash | AutoV2 9F94EEF54C |
This lora was made as a practice for my first try on kohya-ss. It took quite a while for me to understand what I doing and I'm still making mistakes now. But I do like that I'm able finally trial and error with keywords, steps etc. I want to learn more stuff, but right now, that's all I can process...
What does it do?
It generates a fashion style resembling cybersigilism aesthetic, it may work on non-fashion prompt too. This lora loves horns/halo, wings, chains, spikes, ruffle skirt/dress, armor, metallic, chrome, logo, simple/plain background, symmetry (and maybe pink/purple). If you do not want any of these, you can try adding them into your prompting keywords or using negative prompt (it may not work).
What are the triggers?
The main trigger is: Cybersigilism
*For outfits changes, you can explore prompts to custom the "armor" such as navel, shorts, bikini, etc.., you can check out my prompt to help.
Recommended Weight
Depending on the models used, the weight is good from 0.7 to 1.2.
Anything below 0.7 could not utilize this style fully, I had manage to generate results at 1.5, but upscaling will ruin it. Feel free to explore.
Recommended model:
Realistic, 3D and 2.5D model works best to bring out the metallic/chrome material.
Reflection: As a practice/training, I downsize the dataset and the result still as stable as compared to my previous lora trained on civitai. While I thought the lora will be ridge since I kept seeing the "same result" every time, such as a full body front view of a girl, it responds well to new prompt that I introduce to it, proofing itself to be more flexible than expected.