
Troopers! We are green, and very, very mean!
Caution: Your model's coherence and prompt recognition -- especially with military themes -- will affect your results noticeably. Models that can do firearms above-average will give better guns, and some models will resist generating visors, helmets, or even ODSTs at all (variable with prompt). Most of the photographic examples were generated with Juggernaut Final, which performed mostly stably with this LoRA.
This LoRA is still WIP. It kind of sucks. It's partially overfit, partially underfit. And it generates watermarks more than I'd like, though using watermark, signature
in negative prompt helps reduce it. This LoRA was an absolute pain in the ass to train, if I'm being honest.
Nonetheless, I hope to improve on it with dataset tweaks and retraining, then hopefully moving it to SDXL some time, too.
Other tips
Some annoyances can be fixed or reduced with prompting.
Certain prompts just seem to kick you out of ODST-zone, which can usually be solved by describing things differently. The tendency to give bare fingers can be reduced by prompting for gloves, and bare arms
in the negative prompt.
Additionally, when you don't want a weapon, try putting rifle, SMG
in the negative prompt if you find your ODSTs are holding things they shouldn't.
描述:
训练词语: ODST,shemagh scarf,aiming and firing a,battle rifle,assault rifle,SMG,sniper rifle,blue visor,fireteam,squad,pair
名称: ODSTs-beta.safetensors
大小 (KB): 16238
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success