<h2 id="v2.5-description-1e3qm7o9q">V2.5 Description</h2>So I re-ran the training for the model. This time using recommendations by some folks and picked the lowest loss version of the 2nd run, which is epoch-7.<h2 id="original-description-ajejc5ys2">Original Description</h2>Hi! It's been a hot minute.For a bit of background, this model is "Supposed" to be a better trained version of the Luminous Witches style lora. Unfortunately, the model seemed to decide that another target was worth pursuing and went in that direction.<s>It doesn't give off the right style that I hoped for TLDR.</s>Seems like I'm just prompting it wrongly. The style appears to be there but to get a more anime feel to it, try `anime screencap`Since the style is still interesting, I'm release epoch-09.Trigger keyword is "tk-lwshft"

Luminous Witches - V2 Anime Style

Alternative version. This is probably better for most usecases. Trigger word is the same as in the description. Just that it's trained differently.

Files with SHA256: 9b74a653a6126a072f36be792aefcc2f52d81d6162e937b2acc4c1f1618ab17a