Next-Gen AI Generation Preview With Stable Diffusion 3! But with a Danger of Heavy Handed Censorship

[@WelshDragonJS8423-bv7kg]

One of the best things about open source AI is that even if the base of it is censored for a variety of reasons (most of those being ridiculous of course but understandable so the original developers can cover themselves) it won't stop those from creating forks or checkpoints or loras etc that will bypass those.

[@MedicinalSquishing]

Exactly.

[@AI_EmeraldApple]

Yeah the worst case senario is if this is similar to stable diffusion 2.0; which was so crippled no amount of community finetuning by the opensource community was able to make it good. The CEO of stability AI confirmed that SD3.0 uses the same/similar data set as SDXL and stable cascade, which can do NSFW 😃

[@Rohan3CAV]

AI needs to be ripped out of the hands of these censoring soy nerds

[@AI_EmeraldApple]

Yeah I agree 💯💯💯

[@AustinGlamourPhoto]

SD is open source, doesn't matter how censored it is, people will be able to train anything into it. There are 1000s of custom SD models people have trained. Leaving certain questionable things out of the base model is plausible deniability; let the community add the questionable stuff, stability AI didn't do it.

[@AI_EmeraldApple]

Yes this is a good point. The problem previously was stable diffusion 2.0 and 2.1 which was crippled to uselessness and no amount of fine tuning was able to make it work. SDXL now as a lot of uncensored models