I originally planned on this model being capable of being a complete anthro/feral slider, and that is still the plan however, the current version seems to make ferals at high strength and turn completely off at negative strength. this can be avoided if you prompt for "anthro" but I don't really like my models having an "activation token" unless they are designed to, and for this one, I think it is undesirable to have an activation token.
This model was trained on Indigo Furry Mix Hybrid v90. This model was trained with absolutely no images. right about now you might be wondering "What do you mean no images?" Allow me to explain, I used a training framework called "LECO" or Lora for Erasing COncepts. LECOs are significantly more useful than they might initially sound, with the framework, I can also train the model to alias or enhance concepts with sliders, and that's exactly what I did. during training, I set it up to add "feral" and remove "anthro" from every generation, this has the effect of conditioning the model to produce anthros at negative weight and produce ferals at positive weight.
Speaking of training, training models is expensive and I run training on my own private server, if you like what I do, consider supporting the development here!
Most furry models already understand the concept of a feral quite well, so why would I ever need to distill that knowledge into a LoRA? Well, I personally prefer sliders for a lot of concepts, if something is looking a bit too feral, or a bit too anthro, I can change it with a slight tweak to the LoRA instead of rewriting the prompt, or making silly adjustments to the weight of tokens that make the model unstable. additionally, this approach preserves concepts that are typically very closely tied to one another, so you can quite easily make a feral or anthro pony from my little pony, even though it may be hit and miss even when you do prompt for it directly. and if it isn't working, you can increase the weight to an insane degree.
This model currently doesn't operate entirely as I would hope, I want the model to produce anthros at negative weight, and ferals at positive weight however, that is not yet functional, I have been able to emulate that function by prompting for "anthro" always, but I see this as a limitation, not a strength. the hope is that a future version will require no prompting, and be transferrable to hyper-realistic models like absolute reality
Now works on both EPS and Vpred models
trained with my new disentanglement technique.
trained on a diverse set of species so it should perform better even with more niche species.
1. Modèle partagé uniquement à l'apprentissage et au partage. Droits d'auteur et interprétation finale réservés à l'auteur original.
2. Auteur souhaitant revendiquer le modèle : Contactez officiellement SeaArt AI pour l'authentification. Nous protégeons les droits de chaque auteur. Cliquer pour revendiquer
