Type | |
Stats | 161 507 |
Reviews | (27) |
Published | Sep 20, 2024 |
Base Model | |
Training | Steps: 3,000 Epochs: 15 |
Usage Tips | Clip Skip: 2 Strength: 0.4 |
Trigger Words | spot color selective color (black background), (monochrome, grayscale skin, selective_coloring) |
Hash | AutoV2 C4D032B150 |
There is a much more stable version that I only found out about later: https://civitai.com/models/677340/monochrome-neon-glow-style
It is much easier to work with.
===========================================================
LoRA strenght and tags required depend on the checkpoint used!
(Try to avoid monochrome if possible, as it massively reduces saturation)
In gallery you can find examples of different prompts depending on a checkpoint, so if you want to try I'd recommend chosing a pic you like and starting with prompts/checkpoint from it.
So...I tried making a LoRA for the frist time. It...kind of works? I guess? I...think?
(Only later did I realise I could've gone for "grayscale skin" instead and use it as concept instead of style)
Purpose for the LoRA was to help me with producing sound-sensitive wallpapers with HUE shift. Example
For 1024 txt2img It needs a strenght of 0.15-0.7 (the more prompts you use the more strenght is needed)
As for img2img and detailers....still working on that. Putting LoRA strenght at anything above 0.5 seem to reduce quality by a lot there (so far best detailer results at 0.2 and img2img at 0.35).
First version is more of a test than an actual upload.
Used only 20 images to train a style LoRA... Once I recover Buzz I'll try once more using .... 1000? Honestly I don't know what I'm doing :D