Negative prompts are a really interesting as someone who uses LLMs much more than diffusion models. With ChatGPT, for example, you don't really get the concept of a "negative prompt." Saying "avoid X" can sometimes get you what you want, but it can also easily backfire since you've now included the word "X" in the prompt and the AI is more likely to drift in that direction.
Exactly. Diffusion models suffer from the same problem, which is also why negative prompts are added as their own separate field or parameter. Because if you simply write "Boy without a jacket" or "Boy that has no jacket," Midjourney may well focus on the jacket itself and give you more pictures of boys in jackets, ignoring the "no" part. Having the "--no" parameter as a standalone function is a way to solve the issue.
Negative prompts are a really interesting as someone who uses LLMs much more than diffusion models. With ChatGPT, for example, you don't really get the concept of a "negative prompt." Saying "avoid X" can sometimes get you what you want, but it can also easily backfire since you've now included the word "X" in the prompt and the AI is more likely to drift in that direction.
Exactly. Diffusion models suffer from the same problem, which is also why negative prompts are added as their own separate field or parameter. Because if you simply write "Boy without a jacket" or "Boy that has no jacket," Midjourney may well focus on the jacket itself and give you more pictures of boys in jackets, ignoring the "no" part. Having the "--no" parameter as a standalone function is a way to solve the issue.