Text Prompts
Prompts are the pieces of text you enter to instruct the AI model what to generate.
To some it is an arcane art, and there are many opinions and mystic invocations floating around. Here we won't go into that, but will simply describe some things you can do in the Krita plugin which aren't obvious at first glance.
Generally text should be in English and contain letters, numbers, (space or new line), . (period) and , (comma).
Most other things are either ignored, or have special meaning.
Negative prompts can be used to avoid certain content in images. It's not a straight-forward mechanism though, and using it effectively requires some understanding of what it does on a technical level. It is not exposed in the plugin's docker by default, but it can be enabled in Interface settings.
Text from the Style settings is merged with the text entered for the current document. This is useful to avoid repeating general trigger words for content type and quality.
Usually text from the Style settings is appended at the end. The {prompt} placeholder can
be used to indicate the position where the document's text should be inserted.
You can increase or reduce the weight of individual words or phrases in your text. You do this with brackets.
Examples:
| Prompt | Explanation |
|---|---|
a violet rose an a rooftop terrace at (dawn) |
emphasize dawn (weight 1.1) |
a violet rose on a (rooftop terrace:1.3) at dawn |
emphasize rooftop terrace (weight 1.3) |
a (violet:0.8) rose on a rooftop terrace at dawn |
de-emphasize violet (weight 0.8) |
Tip
Use Ctrl + Up/Down (Arrow key) in the text editor to increase/decrease the weight of the word under the cursor. Also works for selections.
LoRA models can be added in the Style settings. But they can also be used with a special syntax directly from the text input.
Examples:
| Prompt | Explanation |
|---|---|
banana <lora:food_photography> |
load the food_photography.safetensors LoRA (weight 1.0) |
apple <lora:food_photography:0.5> |
load the food_photography.safetensors LoRA (weight 0.5) |
Note that the LoRA directives can appear anywhere in the text, it doesn't matter. They will be removed from the final prompt.
You will also get auto-completion help when typing to make it easier to select the correct file.
To use languages other than English for text prompts, automatic translation can be activated in Interface settings.
This will show an indicator with the selected language mode in the text input box.
- Click the indicator to temporarily disable translation. It will show
ENto indicate text is expected in English. - Ctrl + Click the indicator to immediately translate and replace the current text. (Requires local server)
By default only text entered in the text box for documents is translated:
- ✓ includes text for regions
- ✓ includes negative prompts if they're enabled
- ✖ does not include any text from Style settings!
Sometimes you might run into a situation where there is a mixture of text that needs translation, and specific phrases, tags or instructions which are already in English and should not be touched. For full flexibility, the language can also be indicated directly in the text. This still requires translation to be enabled in general.
Language directives use the syntax lang:XX, where XX is a 2-letter language code. Example:
lang:en origami paperwork, lang:zh 狐狸和鹤, lang:en mountain view
This would result in the following text to be sent to the model: origami paperwork, Fox and crane, mountain view
Technically you can have as many directives as you want, and combine multiple languages.