Finally got Illustrious version released. Use that or Pony V2 for best results.
Main appearance tags: mermaid, mermaid tail, head fins, green hair, purple eyes, purple tail, ahoge, side bangs, hair over breasts, large breasts
Clothing tags: black t-shirt, red skirt, pleated skirt, shell necklace, bracelets, purple headband
Optional: gills, scales
Since the dataset is a little more diverse for V2 and Illustrious, you should be fine using them at 0.8 weight, but feel free to experiment. I still don't recommend you go above 0.6 on V1, however. As for anima, I highly recommend you use it at 0.2 to 0.4 if you want any kind of decent posing. Trust me, it'll be able to retain the key features of her design at these weights, I've tested it thoroughly.
Note that the wan video version of the model is highly unstable and somewhat prone to deliver results that may be subpar. I advice to use it in tandem with img2video if possible in order to give it a good origin point.
Description
I might be a bit obsessed with the quality of my model. I modified the dataset once more to remove the poses that looked particularily wonky and were prone to causing wonky poses in the final data, and added some generated images of her that used unique styles and environments, as well as various angles. This was crucial for the training of this version because it prevented the biggest issues I had with previous iterations, that being the forward pose bias, the kisekae style poisoning, and the aggrevating habit the models had to completely ignore background tags at higher weights. That shouldn't be an issue anymore, and I can safely say that it should work fine at weight 0.7 and above now without sacrificing style and posing compabilities, but it might also depend on the outfit you intend to use. Either way, I doubt I'll train another version again soon unless another big issue should come up.