Verified: 5 months ago
SafeTensor
Type | |
Stats | 198 265 |
Reviews | (29) |
Published | Jul 23, 2024 |
Base Model | |
Training | Steps: 818 Epochs: 14 |
Usage Tips | Clip Skip: 1 |
Trigger Words | justin_sweet_style |
Hash | AutoV2 D240C36B0E |
Created on Civitai
trigger word: justin_sweet_style
This is my second attempt at a LORA (with the remaining buzz I was able to spare 😅). In this one I tried to recreate the style of Justin Sweet Illustrations (inspired by one of my favourite games, Icewind Dale, which he illustrated, besides other works). I used 150 images, from Sweet's illustrations, to train it (fantasy, sci-fi, the jungle book)
I was hoping for a good style for fantasy.
Testing it, I find that between 20 and 25 steps and fewer negatives is What gives best results (or at least, closest to the expected style)
Sorry, noob here.
If anyone cares to try this, then I appreciate it.