Tagging
The team seized the opportunity to fill a gap in existing Image Models knowledge and teach the AI about tags! Tags lend much more control to your prompting and allow for consistency across different generations.
When you type your image generation prompt, tagging suggestions will appear.
The completely retrained NovelAI Diffusion allows you to give the AI clear instructions on what to generate. Since the AI was trained with specific tags, it will be able to recognize labeled definitions considerably better.
Tag Knowledge Indicators
The AI will suggest tags based on what you type, and display corresponding markers indicating how much knowledge it has of each tag.
The use of known tags is not required, but they can play a critical role in steering the AI in the direction you want. This will create the illusion that all your characters are consistent across different generations.
Random Prompt
If you can't think of anything specific to prompt for, you can always use our new Random Prompt button!
It will automatically pick from a list of tags for you and create a prompt for you. After the Random Prompt is created, you can generate from the prompt, delete it ⨉ (reverting the prompt back to your previous manually written one) or reroll it ↻ into another Random Prompt.
The Random Prompt is also fully selectable and editable.
Goose Tip: Deleting a Random Prompt will revert the prompt to whatever you were writing before, even if you leave the page. So don't be afraid to test it out whenever you fancy!
Tag Ordering
For our newest NovelAI Anime V3 model, the order of the tags is used to dictate how important a tag is to the final results you'll get.
The earlier a tag appears in the prompt, the stronger its effect should be.
This works in alongside the {strengthening} and [weakening] vectors.
On another note, some tags are 'expected' to be placed in a specific order, at the start of the prompt.
The expected pattern is:
1boy, 1girl, characters, series, everything else in any order
Goose Tip: Note that this is just a recommended order of some common tags.
You can always omit those tags or place other tags before them so that they have more relevancy.
Special Tags
Our models have unique tags made to better define the final aspects of your generations.
There are Quality Tags, Aesthetic Tags, and Year Tags.
Quality Tags
The quality tags are used to influence the overall quality of the generated images.
They are:
best quality
amazing quality
great quality
normal quality
bad quality
worst quality
Here's an example of an image generated with the same settings and prompt, except the one on the left uses the best quality and the one on the right uses the worst quality tag:
Aesthetic Tags
There are also tags that make the generated images more or less aesthetically pleasing.
They are:
masterpiece (V4.5 only)
top aesthetic (V4 only)
very aesthetic
aesthetic
displeasing
very displeasing
And here's an example like the one above. These ones using the very aesthetic and very displeasing tags respectively.
Year Tags
In addition, we also have tags to indicate years which the AI base its art on.
They are:
year XXXX
"XXXX" being whatever year you want to use.
For example, using the tag year 2014 will make your generated images' artstyle to be more in line with the prevalent style of the year of 2014.
Any year should work, although with mixed results.
Renamed Tags
Due to how the syntax of the image generation prompts works, the | character is used as a prompt mixing prompt separator.
Because of that, and because some tags were deemed abstract, these tags were renamed to the following more descriptive names:
v should instead be written as peace sign
double v should instead be written as double peace
|_| should instead be written as bar eyes
\||/ should instead be written as open \m/
:| should instead be written as neutral face
;| should instead be written as neutral face
eyepatch bikini should instead be written as square bikini
tachi-e should instead be written as character image
Other Tags
There are also some other miscellaneous tags in our dataset that don't fit any of the previous categories.
location works like a combination of the indoors and outdoors tags and indicates that your image should show some kind of location without having to specify which of the two it should be.
Note:
Please keep in mind that the NovelAI Diffusion models will not produce the same results as the standard Stable Diffusion model, even with the same prompt and seed.