UPDATE: Model V3 is a retraining of the model with a higher-quality data set, more careful tag pruning and more steps, with the end result more closely resembling the original comic artwork at 0.8-1.0 weight at Clip Skip 2. Requires more weight adjustment than V2 to play nice with other LoRAs as a result, though. Again, if you're having trouble with the outfit, add the positive tags witch hat, orange hair, long hair, aqua dress, black turtleneck, long sleeves, black pantyhose to ensure you get her default look. This version also tends to make her hair very long, so add very long hair and extremely long hair to the negatives if you're having trouble with that.

--

Based on the main character of the Belgian comic series of the same name. Trained on Anything 4.5 with anime.vae, using 70 pieces of art from the original artist, as well as a few pieces of fanart to fill some gaps.

Works well at 1.0 weight, but might require some adjustment depending on what other LoRAs you use with it, some get overpowered by the thick lines inherent to this one. If you're having trouble with the outfit, add the positive tags witch hat, orange hair, long hair, aqua dress, black turtleneck, long sleeves, black pantyhose to ensure you get her default look.