I did my best.

Generating images of Tails with this model is feasible, but it seems a bit difficult.

Without the help of the ControlNet plugin, any inappropriate tag or tag with inappropriate weight can ruin the picture. I tried to build the picture with a lot of tags like other models - the result was a complete failure, and even the phenomenon of the model ignoring the tags, even the simplest 1boy and solo.

Another serious problem is that the Tails drawn by this model will switch back and forth between 3D and 2D styles. And if you write “3D” in the negative tag, the picture is likely to become not very good.

Then, Tails’ fur will also switch between gold and dark orange, dark orange is similar to Tails’ color in “The Adventures of Sonic the Hedgehog”.

Well, considering the training material problem, I guess I can understand this…but it does make me feel a bit uncomfortable.

The images below are all obtained by large-scale batch production with a small number of tags (like in “lottery”!). Some are based on the original image and then assisted by the ControlNet plugin. In short, I feel like I’m drawing cards in a mobile game. Probably 2 or 3 out of 24 generated items look good.

This model can indeed generate very cute Tails, yep, I admit it. But the success rate is a serious problem. Of course, even if you succeed in generating very cute Tails with it, you will still find some minor errors, and in the end you will not escape using ControlNet and PhotoShop…

But I'm still trying to figure out how to properly apply a lot of tags in this model. Maybe I just haven't found the really right way to do it, maybe this model isn't quite the same as the others, maybe...

But in any case, you can't find a better model than this that can draw Tails without the help of a lora model.