The Evolutionary Poster Composer

The Evolutionary Poster Composer is a set of generative experiments that employ Evolutionary Computation (EC) approaches for generating, evolving and evaluating typographical posters. The main motivation behind these experiments is to develop a generative process that emulates the workflow of traditional letterpress print-houses in the 19th century. Letterpress is a printing technique that became popular on the follow-up of the Industrial Revolution because it allowed a cheaper, easier and faster printing of commercial posters for mass communication. However, the design process of these posters was slightly different from the typical contemporary poster design process. At the time, letterpress designers created posters while trying to fill a matrix, often in collaboration with the client. The design decisions of these designers were very pragmatic: extensive sentences were composed in condensed typefaces, and short sentences were composed in extended typefaces. Also, the most important parts of the content were emphasised using bigger typefaces.

We are interested in replicating the letterpress process computationally with three main research and conceptual aims in mind. First, the letterpress design process is by itself an algorithmic process (i.e. the designers’ crafts posters through the use of the same “recipe”) and participative (i.e. the client takes an active role in the definition of the final output). Thus, this approach has the potential to allow the study of how computational design techniques may expand the tools and automate some processes in Graphic Design practice, creating novel ways to communicate with people. Second, the letterpress became popular due to technological and social advances promoted by the Industrial Revolution. Currently, we believe that the democratisation of digital technologies and Artificial Intelligence will have a similar, or greater, impact on the future practice of Graphic Design, changing their production process. Finally, the letterpress posters are artefacts for mass communication and, therefore, they had an influential role in the definition of the contemporary social nature of posters. However, in these experiments, we explore posters as media for communicating personal information, a relatively unexplored subject that goes against this nature of letterpress posters and promotes the discussion about the future of the posters themselves.

These experiments employ EC approaches as the main generative engine. EC approaches can encapsulate the basic set of tasks and activities that often graphic designers perform during their working hours. This way, these approaches may be extremely useful tools for graphic designers, especially for visual and conceptual exploration, in the most exploratory stages of their projects. Briefly, the system employs a Genetic Algorithm to interactively arrange a certain text, divided into text boxes, in order to fulfil a predefined canvas divided into a grid. The text content may be inputted by the user or, alternatively, gathered automatically from multiple web sources. After, a population of candidate solutions (or posters) is evolved using variation operators (i.e. recombination and mutation). The system evaluates the outputs generated automatically. Each generated poster is evaluated considering three objectives: legibility, i.e. if the content on the poster is fully readable; aesthetics, i.e. if the poster is visually appealing; and semantics, i.e. if the visual characteristics of the poster convey the semantic meaning of its content. Natural Language Processing techniques are employed to semantic analyse the text, recognising the emotions and colours related to the posters’ content.

Although the system automatically assigns the fitness of each individual, the users may guide the system’s evolutionary process expressing their visual and conceptual intents (see Figure 2). Currently, we are experimenting with three distinct automatic experiments. Each experiment a different fitness assignment strategy: (I) a Multi-criteria Hardwired Fitness Function; (II) a Multi-objective Optimisation Approach, and (III) a Hybrid Strategy that combines features from the previous two strategies.

Multi-criteria Hardwired Fitness Function Strategy

The Multi-criteria Hardwired Fitness Assignment Strategy considers that quality posters are those that fully satisfy all the evaluation objectives. Therefore, this strategy considers that the quality of a poster is measured by the weighted arithmetic mean of the objective evaluation. The overall merit of each poster is defined by the weighted arithmetic mean of these three objectives. Also, the users may express their preferences by specifying the intended visual features for the output designs, selecting the preferable fitness assignment strategy, and controlling different aspects of the evaluation strategy and objectives evaluation.

Earlier experiments explored the different criteria on evaluation: (I) composition, i.e. how the poster composition fulfils the whole space on canvas; (II) visual design, i.e. how content is possible to read on the poster; and (III) typography, i.e. how much the typographic decision of system please a set of typographic preferences of defined by the users (see Figure 5). You may find some generation examples of this experiment here. Aligned with this earlier experiment, we also developed the Digital Poster (De)Composer, a system when we used this evolutionary approach and, after, we misshapen the outputs using a basic pixel deformation algorithm (see Figure 7).

Multi-Objective Optimisation Approach Strategy

The Multi-Objective Optimisation Approach Strategy considers that quality posters are those that promote more innovative and distinctive relationships between the three objectives (either optimisation, minimisation or balance). The fitness of each individual is defined by the implementation of an Elitist Non-dominated Sorting Genetic Algorithm (NGGA-II). The rank of the non-domination front, where the individual is placed, is used as a quantitative value of the fitness of the poster. The crowding distance is also considered in the selection of the individuals during the tournament. Also, the users may define the objectives that they want to optimise.

Hybrid Strategy

The Hybrid Strategy considers that quality posters should always be readable and present innovative and distinctive relationships between the semantics and aesthetics objectives (either balance or optimisation/minimisation). Thus, the strategy evaluates each poster by calculating the weighted arithmetic mean of legibility measure (the hardwired part) with the relation between the semantics and aesthetics objectives (the multi-objective part). The relation between the aesthetics and the semantics objectives is computed similarly to the multi-objective optimisation approach strategy; however, it only uses two objectives. Crowding distance is also used in case of a tie in the tournament.

Resources

Colour and Typography Setting Sheets
Example outputs
Example outputs (content gathered from Twitter)

Related

The Posters’ Factory
Evolutionary Poster Composer at xCoAx 2018

Exibhitions and Installations

Publications

• S. Rebelo, C. M. Fonseca, J. Bicker, and P. Machado, “Evolutionary Experiments in the Development of Typographical Posters,” in 6th Conference on Computation, Communication, Aesthetics & X, Madrid, Spain (xCoAx 2018), 2018, pp. 65-75.

  • Bibtex
  @inproceedings{rebelofonsecaBM2018,
  Author = {Rebelo, S\'{e}rgio and Fonseca, Carlos M. and Bicker, Jo\~{a}o and Machado, Penousal},
  Booktitle = {6th Conference on Computation, Communication, Aesthetics \& X, Madrid, Spain (xCoAx 2018)},
  Editor = {Rangel, Andr\'{e} and Ribas, Lu\'{i}sa and Verdicchio, Mario and Carvalhais, Miguel},
  Pages = {65-75},
  Publisher = {Universidade do Porto},
  Title = {Evolutionary Experiments in the Development of Typographical Posters},
  Year = {2018}}
  • |
  • PDF

• S. Rebelo, J. Bicker, and P. Machado, “Evolutionary Experiments in Typesetting of Letterpress-Inspired Posters,” in Proceedings of the Eleventh International Conference on Computational Creativity, September 7–11, 2020, 2020.

  • Bibtex
  @inproceedings{rebelo2020a,
  Author = {S\'{e}rgio Rebelo and Jo\~{a}o Bicker and Penousal Machado},
  Booktitle = {Proceedings of the Eleventh International Conference on Computational Creativity, September 7–11, 2020},
  Title = {Evolutionary Experiments in Typesetting of Letterpress-Inspired Posters},
  Year = {2020}
  }