Are questions about specific AI art packages on topic?

Question

Is it on topic to ask a question about a specific AI art package. For example, to ask how Midjourney handles certain prompts, or do we need to keep the questions generic and non package specific?

Answer 1

I don't think this needs careful deliberation. Yes, questions about specific AI art packages are on topic. We've been asking such questions since the early days. I've asked two myself (here and here).

For example, to ask how Midjourney handles certain prompts...

On-topic.

... do we need to keep the questions generic and non package specific?

Also on-topic. (Although may be closed for other reasons, like being unclear or too broad.)

@Wicket, some stacks will automatically delete any questions regarding named products or brands, and will only allow you to post a general question about a class of products. I'm curious as to whether this stack is one of them.

Feel free to name the service you're using in questions (it'll help users write more useful answers) and answers (it'll help the reader replicate what the answer says).

Answer 2

I am not familiar with the term "packages" here that you used. I will interpret it as GenAI tools like ChatGPT, Midjourney, and DALL-E. Feel free to correct me if I'm wrong.

Is it on topic to ask a question about a specific AI art package?

You can ask about a specific GenAI tool. However, it essentially depends on the topic of the question whether the question is going to be on-topic or not.

How Midjourney handles certain prompts?

I interpret this as the question is asking about the internal workings of generative AI model(s), which doesn't fit within the scope of this site; hence, off-topic.

Do we need to keep the questions generic and non package specific?

Not necessary. You can keep the questions generic if you want. If you have questions regarding any specific GenAI tools, you can also ask about it. Be sure to tag the questions accordingly.

Answer 3

If by "package" you mean model, then I fear this is a defacto requirement and part of the issue I had described before. You will have to make tags for every single model, and often specify the version too.

Forgive me for referring to Stable Diffusion specifically but I think this can safely be extended to other systems.
When writing a SD prompt, the model heavily influence how the prompt should be written.

Different models interpret the same word differently: as a simple example, many Anime centered models seems to use a tagging system that originates in image database sites (see the various "booru"). As such you end up with things like:

1girl, long hair, sitting, outdoor,

in this context 1girl doesn't seem to carry any "age" related meaning.

Knowledge of concepts is also relevant. If you use PonyDiffusion you could be able to write:

Fluttershy, human, Equestria Girls

and get a decent enough result. Try that on base stable diffusion and you will get a mess that only vaguely has some Fluttershy element. Try on Dungeon&Diffusion and you will get nothing even close. Gets even more complicated when you start referencing less known characters.

Some models are generic, some have a specific aesthetic. On some model you can get different styles by writing:

... by Artist.

Some others ignore that, produce results that does not seem to match the artist or use different artists.