Type | |
Stats | 217 81 15 |
Reviews | (44) |
Published | Sep 6, 2024 |
Base Model | |
Usage Tips | Clip Skip: 2 Strength: 0.8 |
Trigger Words | inueShinsukeStyle, |
Hash | AutoV2 2374675FB0 |
(for bounty)
Main trigger: inueShinsukeStyle,
A lora for the style of Inue Shinsuke.
I am experimenting with optimal style settings so there will be two versions to see what people like more/prefer to use:
(BEST SO FAR)Soft: less style adherence, more concept recognition and flexibility
(this one actually worked better than expected, so it will be the main version for now)
From preliminary testing I would recommend starting with a strength of around 0.8 and then tweaking according to preference.
(FAILED)Hard: Stronger style focus, ignoring training on anything else
(this one did not come out as expected unfortunately, I will still post it soon for science as version 0.1H(UPDATE: NOW POSTED AND LIVE), but from my testing so far it is basically worse in every way and I do not recommend for anything other than experimentation for lora related data collection or if you are really really bored... OR if you actually like the slightly different and weaker style guidance it provides)
Recommended strength at 1.0 for this one, and maybe even higher.
(PLANNED)New: This is a version I am thinking about making with a reduced dataset of just the newest pictures from the style to capture that specific feel. The image dataset would be smaller than recommended for a style lora but I am tempted to try. It is on my to-do list but pretty far down so do not expect it anytime soon...