I disagree.
Using tools such as Stable Diffusion is like asking a malicious compliante genie a wish.
You have to account for all the "bad" things and use negative prompts for the things that you don't want, eg: "ulgy, deformed, extra limbs"
The aberrations don't really need to come from the training data itself, but are produced when content is generated with it.
I disagree.
Using tools such as Stable Diffusion is like asking a malicious compliante genie a wish.
You have to account for all the "bad" things and use negative prompts for the things that you don't want, eg: "ulgy, deformed, extra limbs"
The aberrations don't really need to come from the training data itself, but are produced when content is generated with it.