The Dead (Furry) Internet
Feb. 21st, 2026 09:37 am![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
elfsHenry Farrell (![[bsky.social profile]](https://www.dreamwidth.org/img/silk/identity/user_other.png)
himself) asserts that LLMs are “cultural technologies,” stores of cultural information with patterns of retrieval that are new and to which we will adjust. If this is true, then Furry Fandom should probably be doing a lot of introspection and undergoing a lot of change.
Let’s emphasize that nothing I say here is the fandom’s fault. I am not blaming anyone. If I’m going to indict anyone, it’s the people who make the Furry-oriented models for Stable Diffusion and other image generation software, because that’s where the critical fault lies.
It’s an open secret among the people who enjoy porn illustration renders (not gonna call it “art,” it never is) that the best way to get a really good porn pic is to render it using a furry model first, and then use the image-to-image feature to say, “Rerender this with humans.” The furry community has an obsession with categorizing its interests, and the people who draw lewds will draw and tag everything. The accurate and comprehensive labeling on E621 can be found embedded deep in any furry-oriented rendering model, exceeding by orders of magnitude the accuracy and precision of models based off other image sites like Rule34 and its ilk.
It was also a commonplace joke that the first generation of rendering models, the Stable Diffusion Series 1 models, were incredibly bad at rendering men. It sometimes seemed as if SD1.5 and its offspring had only ever seen one penis, and an exceptionally ugly one at that. The furry models, on the other hand, have an incredible catalog of male bodies and body parts, and to this day exceed the state of the art in some of the models that came after Series 1.
The furry artist community is being devastated by its own success. The thieves (and they are thieves) who dumped all of E621 and other furry art collections into the first generation of furry illustration models, and who collectively managed to create models better tagged and far more comprehensive than the commercials ones created by Stable Diffusion, Midjourney, Dalle-E, and NovelAI as a non-commercial, under-the-radar operation, exploited this otherwise admirable dedication to accuracy to wreck the furry art community’s future.
At the same time, that labeling has gone on mostly in terms of acts and artists, not characters. If LLMs are stores of cultural information, then in thirty year someone will still be able to “render this picture in the style of Personalami, Chunie, and Taran Fiddler,” but you won’t be able to call up any original characters that don’t belong to a massive franchise. Thirty years from now you’ll be able to render (and render porn of) Lola Bunny, Maid Marian from Disney’s Robin Hood, or Rocket Raccoon, but the primary original characters from the founding of the fandom are completely lost. There are no skiltaire in these models; no Omaha the Cat Dancer; no Erma Felna; no Hervystia; no one from Doug Winger’s fertile (in many senses of the imagination) cast of characters; no one from Ken Cougr’s stable.
Maybe that’s a good thing, but if LLM’s are going to be with us for a while, and they’re “stores of culture,” then they’re admissions that Furry Fandom, for all its pride in wanting to be non-commercial, does not love its own artists nearly as much as it loves Paw Patrol (of which, at last count, there were 103 characters encoded in a variety of open-source illustration models).
himself) asserts that LLMs are “cultural technologies,” stores of cultural information with patterns of retrieval that are new and to which we will adjust. If this is true, then Furry Fandom should probably be doing a lot of introspection and undergoing a lot of change.Let’s emphasize that nothing I say here is the fandom’s fault. I am not blaming anyone. If I’m going to indict anyone, it’s the people who make the Furry-oriented models for Stable Diffusion and other image generation software, because that’s where the critical fault lies.
Obsessive Labeling
It’s an open secret among the people who enjoy porn illustration renders (not gonna call it “art,” it never is) that the best way to get a really good porn pic is to render it using a furry model first, and then use the image-to-image feature to say, “Rerender this with humans.” The furry community has an obsession with categorizing its interests, and the people who draw lewds will draw and tag everything. The accurate and comprehensive labeling on E621 can be found embedded deep in any furry-oriented rendering model, exceeding by orders of magnitude the accuracy and precision of models based off other image sites like Rule34 and its ilk.
It was also a commonplace joke that the first generation of rendering models, the Stable Diffusion Series 1 models, were incredibly bad at rendering men. It sometimes seemed as if SD1.5 and its offspring had only ever seen one penis, and an exceptionally ugly one at that. The furry models, on the other hand, have an incredible catalog of male bodies and body parts, and to this day exceed the state of the art in some of the models that came after Series 1.
The furry artist community is being devastated by its own success. The thieves (and they are thieves) who dumped all of E621 and other furry art collections into the first generation of furry illustration models, and who collectively managed to create models better tagged and far more comprehensive than the commercials ones created by Stable Diffusion, Midjourney, Dalle-E, and NovelAI as a non-commercial, under-the-radar operation, exploited this otherwise admirable dedication to accuracy to wreck the furry art community’s future.
And A Loss of History
At the same time, that labeling has gone on mostly in terms of acts and artists, not characters. If LLMs are stores of cultural information, then in thirty year someone will still be able to “render this picture in the style of Personalami, Chunie, and Taran Fiddler,” but you won’t be able to call up any original characters that don’t belong to a massive franchise. Thirty years from now you’ll be able to render (and render porn of) Lola Bunny, Maid Marian from Disney’s Robin Hood, or Rocket Raccoon, but the primary original characters from the founding of the fandom are completely lost. There are no skiltaire in these models; no Omaha the Cat Dancer; no Erma Felna; no Hervystia; no one from Doug Winger’s fertile (in many senses of the imagination) cast of characters; no one from Ken Cougr’s stable.
Maybe that’s a good thing, but if LLM’s are going to be with us for a while, and they’re “stores of culture,” then they’re admissions that Furry Fandom, for all its pride in wanting to be non-commercial, does not love its own artists nearly as much as it loves Paw Patrol (of which, at last count, there were 103 characters encoded in a variety of open-source illustration models).
