>> CHECK OUT PONY DIFFUSION V5 <<
Pony Diffusion is a Stable Diffusion model that has been fine-tuned on high-quality pony, furry and other non photorealistic SFW and NSFW images.
This model is capable of producing SFW and NSFW content so it's recommended to use 'safe' prompt in combination with negative prompt for features you may want to suppress (i.e. nudity) if you want to avoid NSFW outputs.
Despite its name, this model is capable of producing wide range of furry and cartoon images as side effect of improving data diversity (with exception of anime styles, for which Waifu Diffusion is much stronger choice).
Join our Discord Server to get access to new versions of this model and free Stable Diffusion bot.
Special thanks to Waifu-Diffusion for providing finetuning expertise and advising through the process, without their help this project would not exist.
Description
Pony Diffusion V4 has been trained on 3M pony, furry, cartoon and anime images filtered based on human scores.
FAQ
Comments (6)
Where can i find the derpibooru_p_95 embedding
I don't think it's an embedding, I think it's an internal tag the model is trained on. When I prompt with it the style is more consistent.
yeah i was thinking that too, it seems all the prompts around that model use it but its actually been inconsistent for me.
i wonder why I'm just NOW finding this. I had searched before and it never appeared....but then I see you only recently uploaded it.
lurk more
if this model is trained from scratch then is there a .CSV file for this model so people can know what tags this model was trained with and so people can search through the tags with the autocomplete instead of having to ask "what tags do i use that are specific to this model?" or "are there specific quality tags so to this version of the model and have they changed from the previous version?"
no sense in having a model trained if people dont know what tags were included or were pruned from the training data, what works or doesnt work or if they were trained like "example_tag" or "example tag".
even if a .CSV doesnt include tags that are specific to what the model was trained on that werent in the original sources, having a derpi (im assuming this is where the training data and tags come from) specific .CSV similar to the stock e621 and danbooru ones for autocomplete would be useful for this dataset.