Update 09.03.2025: V6 (OLT) is out.I still recommend you to stay on V5 (Endgame) as it's the most stable one. First of all, this version was trained on a HUGE amount of new data. Some of it is synthetic data (something I've created with AI and then fed it into training), I did that to make it possible to output concepts that don't exist in real life (look at the preview pictures and you'll see what I mean). A big general-purpose dataset was also thrown into the training to improve some SFW knowledge. With V6 version, less is more. Don't use unreasonably long prompts, try not to use negative prompts at all, use low CFG ~3-4.5.