invoke-ai
Matrices and negative prompts?
Greetings,
Negative prompts would be a nice addition:
great results: https://www.reddit.com/r/promptcraft/comments/xwyi2v/stable_diffusion_negative_prompt_experiments/
inspiration: https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Negative-prompt
What about: portrait of chuck Norris fighting super man | bill gates | sponge bob -n3 result: generates an image with each hero
or alike?
thank you,
Put any number of terms in square brackets to have their concepts excluded from the image. See Negative and Unconditioned Prompts
Personally i'm not a fun of the '|' operator, it make the prompt diffcult to read, also it will non easy to limit its scope. Indeeed, If used at the begining of the prompt or in the middle, we need to introduce some kind of scope delimither, like curly brackets for example.
An alternative idea will be to have a kind of batch prompt, using a template mechanism to fill values in the prompt, I'm thinkig to that.
good point i3oc9i, while "difficult to read" remains subjective, where does the effect of the | ends is quite real. any of these would work but comma delimiter should be just fine: [super man | bill gates | sponge bob] (super man | bill gates | sponge bob) {super man | bill gates | sponge bob} , super man | bill gates | sponge bob,
Thinking about it, having a consensus would be beneficial. The way I discovered about negative prompt was from a posted image with embed prompt in the metadata of the image on discord. To automate the retrieval of the prompt (drag and drop the image into the UI), I have to parse every single convention out there because there is no standard. Just saying, we might want to reach out and coordinate rather then make another standard.
may be comma is not the best option. Thinking to something like this:
An old man walking in the middle of a (forest | city | canyon), photo realistic, in the style of ( author1, author2, author3 )
This prompt should generate 9 images
I guess you meant:
An old man walking in the middle of a (forest | city | canyon), photo realistic, in the style of (author1 | author2 | author3)
That would work beautifully!
`An old man walking in the middle of a (forest | city | canyon), photo realistic, in the style of (author1 | author2 | author3)
Yes, wrting in the metro with a phone it is somethime complicated..... :)
Automatic1111 uses the convention of {1, 2, 3} for matrix prompts, perhaps it's best to stick with that here as well?
Taking into consideration that Automatic1111 is sailing quite nicely among the community. It might be a good idea to mimic it's convention all together. Not super noble but it seems like a smart move.
regarding the negative prompt
i really prefer the way automatic1111 handled this. Have a completely seperate text input box. This makes it very easy to visualize what is negative and what is not
Just checking on progress... is there any hope for | operator in next release?
as for the features listed here
https://discord.com/channels/1020123559063990373/1049656381273292900/1049656381273292900
2: being able to use matrix prompts like a beautiful landscape|during winter|during summer should produce 3 images for each sequence
