Lopunny is one of my favorites, but I had been getting frustrated with how all models that can generate them will force full-furry anatomy on their body instead of capturing their plushie-like shape. This is an experiment to try to remedy that, though I'm still learning how loras interact with models and would appreciate tips from other people with experience.

ANIMA V1 and Illustrious V3 (Lopunny Only)

A completely hand-retagged version based on e621 and booru tags, bigger dataset, and more overall flexibility. Trigger tag is still l0punny , also uses on_model. Due to how many checkpoints use anthro-fied versions of lopunny, you may still have to use a higher strength 1.2 - 1.5 with this lora and put some humanoid features in the negative prompt.

Since Anima is a newer model, I'm seeing what needs to be tagged or trained differently than an Illustrious lora. All helpful tips are appreciated on both fronts. Thank you.

V2 (Lopunny Only / Fewer Tags)

A trimmed / retagged version of the dataset that only features Lopunny and doesn't tag it, instead using Pokemon (creature) and more generalized tags. This is an attempt to make the desired features show up more often, but will likely make it harder to mix with other details.

The trigger tag is now l0punny, which was the originally intended spelling.

V1 (Lopunny / Mega Lopunny)

Full strength 1.0 or higher is recommended, and might not always work without tags like featureless crotch and monobutt with an elevated prompt, but it's been more successful than trying to get the same results with a base model alone.

Use lopunny or mega_lopunny as part of the prompt to get your rabbit. Pokemon_\(creature\) can also be helpful. I can try making a tagless version in the future to see if I can get more consistent results, but it's probably going to require more examples or a smaller dataset first.

The trigger tag is l0ppuny, which will be changed in v2.