Towards the Automatic Customisation of Editable Graphics

In creative fields such as Graphic Design (GD), finding disruptive visual solutions that attract people’s attention is of the utmost importance, either to create communicative or artistic design artefacts, e.g. so the designs stand out over other posters on the streets or other book covers on store shelves. However, as the urgency for more effective designs grows and the GD area is increasingly democratised, graphic designers need to deliver faster and cheaper, which often leads to the adoption of trendy solutions and precludes the exploration of innovative visual solutions.
 

This work proposes a computational approach for automatically styling graphics so that these can relate to given semantic concepts. More specifically, we propose using ConceptNet (Liu and Singh, 2004) to assess the semantic similarity between given keywords (that must be set by the user to describe a given concept) and labels (not changeable by users) given beforehand to a set of mutation methods and their respective parameters. The resulting similarity values are then used to calculate the probability of each mutation method and respective parameters to be used for styling a given graphical item, e.g. given the keyword “sky” a given item may be more likely filled in blue in detriment to other colours. In other words, given “sky”, if the method “fillColour” is picked, the most probable parameter would be blue.
 

Although our approach might be generic enough to be used in different creative contexts, so far it was tested in the generation of posters (see Figure 1), i.e. for transforming and styling a number of text boxes, images, and geometric shapes within 2D pages. Furthermore, to create a tool that could be easily integrated into a designer’s workflow, the presented approach was implemented as an extension for Adobe InDesign — a broadly used desktop-publishing software for GD. Thus, designers can alternate between manually and automatically editing items without leaving Adobe InDesign.
 

The generated posters were evaluated by means of a user survey. The latter suggests that, although the system can be further improved, for the tested experimental conditions, the present approach can be viable to automatically style graphics so that these visually relate to given keywords. Especially, taking into consideration that GD concepts can be transmitted in abstract ways. Evens so, human collaboration might still be essential to curate and fine-tune the results and transform the generated ideas into final GD applications.
 

Publications

References

Liu, H., and Singh, P. 2004. Conceptnet — a practical
commonsense reasoning tool-kit. BT Technology Journal
22(4):211–226