In Gaizauskas et al. (2015), we proposed a definition of Visually Descriptive Language. Intuitively, a segment of text is visually descriptive if we can determine what it says is true or false by visual sense alone.
More precisely, a text segment is visually descriptive iff it asserts one or more propositions about either (a) a specific scene or entity whose truth can be confirmed or disconfirmed through direct visual perception, or (b) a class of scenes or entities whose truth with respect to any instance of the class of scenes or entities can be confirmed or disconfirmed through direct visual perception.
The full definition and annotation guidelines are currently available here.
Our annotated corpus of Visually Descriptive Language is available here. The original corpus is described in Gaizauskas et al. (2015), and is later extended as described in Alrashid et al. (2019) to cover six additional chapters from the Wonderful Wizard of Oz.