Pushing Pixels Through Precise Mishandling

Theodore Davis

This paper establishes a method for the discovery and generation of innovative imagery through the precise mishandling of commonplace digital media.


While the process of generating images in a digital format is nearly the standard in today’s society, personal or direct confrontation with the qualities that compose this medium are rare to none. Through the use of commercial software, which offers tools analogous to hand techniques, the range of visual possibilities is, to no surprise, nearly identical to that of their analog inspirations. Further exemplified by the reoccurring reference to a 'canvas', this selective focus onto the surface of an image greatly ignores the digital code of which the media is entirely composed. As common Internet file formats such as the JPEG/GIF/PNG are used to pass visual information from person to person, one may only notice varying levels of compression to the pictorial content. This remains a passive process to the average user- that is until a glitch occurs along the path of transmission. Disturbing the content, this rarely desired error produces the unpredictable artifact, potentially revealing inherent qualities within that very file format.

Through the advances of error detection and correction techniques, such a phenomenon occurs with less and less frequency, further removing a personal awareness for the medium itself. Therefore what if the glitch was caused intentionally? How might the system react if there wasn’t a stratagem prepared to counterbalance its effect? It is from this inquisition that the paper begins with an inquiry into the properties of a digital image, further exploring the visual and theoretical implication of manipulation through a precise mishandling of the structural code[1].

A precedent for file format manipulation can be found in Cory Arcangel’s 2003 work Data Diaries[2]. Produced in a time of limited error control, he discovered the ability to remove all data from the Quicktime video file format, while maintaining the header used for describing how the video should be rendered. With the original content removed, it defaulted to displaying the computers own daily stored RAM as if it were video data. The results were streaming arrays of pixels, waves, bands, as 128 megabytes of storage were transcribed. This repurposing of media could be considered a form of cross–media visualization, in which, the precise mishandling of documents has the potential to discover innovative systems of representation.


Under the premise that nearly all digital files are speaking the same language of binary code[3], it became plausible to mix such contrasting phenomena as text and image together[4]. In an analog domain, this could be done through the use of collage, physically layering one over the other. However in the digital domain, these elements have the capability to be combined underneath the surface, again on the structural level.

What began as a study comparing and combining texts with contextually relevant imagery developed into a technique whereby the entire pictorial contents are replaced, thus producing a new form of abstraction containing both the text and image as one. Following extensive experimentation with this technique, a particular methodology was then implemented as a web application in order to make it available across every computer platform.

TEXT2IMAGE [www.text2image.org] was designed having simplicity and universal access in mind. The interface consists of little more than an empty textbox and an assortment of colorful images arranged on a grid. Upon loading the website, the textbox is automatically focused upon, ready to accept virtually any textual input from the user’s keyboard. The submit button is pressed and within seconds the main image is refreshed, as a small caption below repeats the text that was entered and an array of recently output imagery is updated. The user has the possibility of switching between two different algorithms, one generating colorful images, the other generating black and white images.

Knowledge of how users interact with the application is observed behind the scenes through a custom set of analysis tools to sort the 60,000+ textual and 120,000+ visual entries, which have been catalogued since launching in February of 2009. While it’s common for the first input to be that of the user’s name, quickly the entries themselves become abstract, adopting a meta-quality, in an attempt to learn how the system works. Thereby leading the user to an active exploration of what resides underneath the surface of the digital image.

With the propagation of TEXT2IMAGE through blogs of varying languages and countries[5], it became clear that the imagery being produced struck a chord within the unseen and unknown. This provided a rich experience, for both the creator and users, witnessing how the overall process and generated images were being interpreted and rationalized- both of which are further explored by the proposed paper.

[1] Manovich, Lev. The Language of New Media (Leonardo. 1st ed. MIT Press, 2001.

[2] “DATA DIARIEZZZZZZ.” 15 Juli 2009. .

[3] Lister, Martin et al. New Media: A Critical Introduction. 2nd ed. Taylor & Francis, 2008.

[4] C. Gänshirt, Tools for Ideas, Basel/Boston, 2007

[5] Andrew Vande Moere , February 2009. information aesthetics, “text2image: Transforming Text into Visual Glitches“ (accessed February 23rd 2009)