Direct Image Descriptions
Last updated
Last updated
Sometimes it's beneficial for Characters to have an image description that is separate from the rest of the text response.
This requires you to set up some examples in your Definition that gives some examples of what this Character cares about when it goes to generate descriptions of images.
The Direct Image Descriptions are always in italics using the ** format. You can also click the "Insert example image description" button in the Character Editor and fill in your example Direct Image Description.
Note that Characters that generate images will always try to generate images on every response. If you design a Character without Direct Image Description examples, or if the Character response ends up not containing any direct (italics) text to use, the system will use fall back to using Inferred Image Descriptions decide what to illustrate this turn.
The Pumba Character generates text that describes the scene and actions (shown in italics below) as well as what Pumba is saying (shown in plain text). With visible Descriptions left on, it shows the descriptive text as well as using it to form the image description.
In Pumba's Definition, a pattern is established that Pumba will "say" things to you, and some action or description is contained in italics after each response (Markdown uses *shown in italics* to indicate what should shown in italics.
If multiple Direct Image Descriptions are generated in a single message, the Character will try to figure out the best way to combine them. We generally recommend only one Direct Image Description per Character message.
