From a basic digital patternmaking such as, for example, a sleeveless shirt or a skirt, the program generates new forms, and is limited only by the pre-established values of deformation, such as a maximum variation of curves or straight lines that compose the patternmaking. Then the software generates a new pattern from these deformations of the original model, which may be printed later in actual size, cut out in fabric and re-built for the wearer’s body in interaction with the program. Each form generated and each garment assembled are the result of a random and unique computer process.

The random element is the main characteristic of creation originated by the work with the software. In the usual process of styling, the desired cut is designed and from the interpretation of this sketch a two-dimensional pattern is made, which will then be assembled in order to reach the closest version possible to the initial design. In this initial pattern, corrections are made so that, in addition to faithfully reproducing the sketch, it also involves the wearer’s body in a precise way. When using Generator as a platform for the creation of a garment, there is no three-dimensional image or a sketch to follow. After the patternmaking is ready, nothing, or very little, can be done in the two-dimensional pattern. That means that the creator, instead of working by creating a design, will work by handling the pattern cut out in fabric on the human body or model and in the resolution of a form, which may be similar or not to a conventional patternmaking, and that should finally involve in some way a human body. The point of interest here is to transfer the creative exercise from one moment to another in the creation process. Because no one has control over the form to be generated, unexpected possibilities emerge, as well as great curiosity in problem-solving (transforming the deformed patternmaking into a real piece of clothing).

The first experiences with the software Generator were transformed into three-dimensional garments and presented in the Rio de Janeiro fashion week in January 2009. However, the format of the software was still at an early stage, so it did not present with a user-friendly platform as the one we intend to distribute to schools and experimental ateliers in the areas of clothing and digital research.

The proposal for Marginalia+Lab was to advance in software design and operation, making its access and handling more accessible to those who are not familiarized with the language of digital data processing. In addition to that, we proposed that the deformation be controlled by an artificial intelligence that would work in transparent ways for users and that would be fuelled by the interaction with them, so that it could learn which deformations users consider most desirable. In a last stage of development, the intention is to provide schools searching for experimentalism and new creation paths with the new version of Generator, using the software as a platform for creative exercises in the design of objects, clothes, and forms.

Implementing the proposal required research on user interfaces and on techniques of artificial intelligence. The graphic interface created allows real time interaction with the forms generated and enables export for printing in a proper cut size, storage in an internal database for future use, among other functionalities. For the software to learn from the interaction with the user, a genetic algorithm was implemented. In a genetic algorithm each individual is classified by an assessment function that determines which individuals are better able to reproduce. In Generator, each deformation generated is considered an individual and the group of deformations already generated is considered the population of the algorithm. Deformations well classified by the user have bigger chances of staying in the population, guiding the software’s artificial intelligence to generate similar deformations in the future. In the same way, deformations classified as unwanted tend to be left out of the population and do not appear in future generations. The non-deterministic behaviour of the algorithm, which may keep less able individuals amid the population instead of excluding them, aims at maintaining the genetic diversity of the population, thus trying to avoid that the evolution, or the choice of forms by the artificial intelligence becomes too limited.

The result of this effort is a new version of the software that can be accessed in this link, where the future development of the program can also be followed.

ABOUT THE AUTHORS

Luis Castilho: Programmer educated in Computer Science at PUC Minas and Computational Maths at UFMG. Has participated as  system analyst in several groups and collectives for about 4 years. Has worked with software development since 2002 and participates in research projects in the area of characterization, patternmaking, and system performance analysis. Currently works with production of visual content for software at the company Coddart.

Julia Valle:Visual communicator graduated from UFMG (Brazil) and a Stylist from the Designskolen Kolding (Denmark), Julia Valle works in several areas of applied arts. She worked alongside great names of the national and international fashion, such as Alphorria (BR), Printing (BR), Henrik Vibskov (DK), and Maja Mehle (SLO) and had her collections shown in galleries in Brazil, USA, and Slovenia. She was selected for the prize Rio Moda Hype four times in a roll. Now working at her own atelier, where she produces and shows her creations in Belo Horizonte and in the Rio de Janeiro brand Redley.