Are you tired of generating low quality blobs with your current stable diffusion model? This finetune is a game-changer in the world of AI art generation, and it’s perfect for creating stunning SFW and NSFW visuals of various anthro, feral, or humanoids species and their interactions based on simple natural language prompts.

Did you know that V6 is actually an upgrade from the previous version? That’s right, this model has been trained on over 2.6 million images aesthetically ranked by its creator (who we assume is a pony), with roughly a 1:1 ratio between anime/cartoon/furry/pony datasets and safe/questionable/explicit ratings. And get this, about half of all those images have been captioned with high quality detailed captions!

Now, how to use Pony Diffusion V6 XL in your AI art generation game. To set the stage: make sure you load the model with clip skip 2 (or -2 in some software) or else you’ll end up with low quality blobs. This model supports a wide array of styles and aesthetics, but provides an opinionated default prompt template that allows for high-quality samples without any negative prompts or other quality modifiers like “hd” or “masterpiece.”

But what if you want to add some specific tags to your prompt? No problem! This model is designed to recognize many popular and obscure characters and series, so just use one of the following templates: `anthro/feral pony, rest of the prompt` or `source_pony, rest of the prompt`. And if you’re looking specifically for pony style, we recommend using either of those two templates.

Now, some technical details. This model has been trained on a combination of natural language prompts and tags, so describing intended results with normal language works in most cases. Using Euler a with 25 steps and resolution of 1024px is recommended, although the model generally can do most supported SDXL resolutions.

But be warned: this model will sometimes generate pseudo signatures that are hard to remove even with negative prompts. This is unfortunately a training issue that would be corrected in future models. If that’s an issue for you, we suggest trying V5.5 or inpainting instead.