BY Thomas Stammer

For our last workshop “Digital Set Extension” at the Baden-Württemberg Film Academy, we chose the film BLADE RUNNER as the thematic basis for an AI-based design. At first glance, the results were complex and detailed. Impressive! At second glance, however, it's pure cliché: Any urban mega-city with futuristic skyscrapers and neon-lit, rainy scenarios. Boring mainstream.

Why is that? The AI generates images from big data archives, which are of course predominantly filled with images from pop culture and mass production. This preference results from the abundance and popularity of such content in the image archives used, which make up the most common "art" images on the Internet. They bring together the preferred content of Western monocultural leisure escapism: gaming, advertising and the omnipresent fantasy image trash. Due to the mass production of AI images, the AI snake also bites its own tail. Because the flood of AI images also becomes part of the image archives, and so the algorithm also devours its own visual waste.

What strategies help us avoid this trash effect when using AI? For example, to gain access to the design of BLADE RUNNER, it is necessary to understand its construction method. Instead of just eating the cake, a chef must know the recipe and then learn how to use the ingredients. In the case of BLADE RUNNER is this the subtle and complex construction of film design. The visual world of this film is constructed from different, sometimes contradictory levels: from Moebius comics, Lawrence G. Paull's retrofitting concepts, the original historical motifs in LA, Sid Mead's 70s futurism and Scott's Ridleygrams [quick sketches with wich director Ridley Scott conveys his ideas to others, editors]. AI cannot recognize or use these construction layers of the final visual product separately.

A hybrid approach therefore makes sense for successful collaboration with AI. Instead of continuously training the AI with more subtle instructions, I use AI material as a building element in the design process. First I use the AI as a kind of smart, interactive image search engine, then I break the results down into individual parts to integrate them into my spatial models. I use these as bases, which in turn are decorated with Photoshop Beta. The core of my current way of working is to use the AI's image production itself as a demolition dump for its own image production. In an ironic victim-perpetrator reversal, the AI images are deconstructed and reused in the individual design process.

Another option is to make the basics of the prompts as specific as possible. Using BLADE RUNNER as an example, I used the original description of the novel as a source. Because Philip K. Dick's descriptions are still largely unexplored territory. In contrast to Ridley Scott's vision of a rainy, overcrowded mega-city, the novel takes place in dusty, flat suburban areas, poisoned and contaminated landscapes with empty, abandoned ruins. On this basis, AI opens up a number of inspiring worlds. This direct implementation of the literature template is a good starting point for your own draft.

One of the biggest problems and obstacles with this method is the censorship of the images by the AI programs. Because certain words are forbidden on Midjourney , including flesh, blood, booty, sex and naked. European freethinkers are not wanted, the AI only allows Anglican-Protestant values of Silicon Valley. This limitation of content is one artistic problem. But film images should not just “look good” and be based on gimmicks or effects, but should involve the viewer psychologically and spatially in the narrative - evoking an emotional and mental reaction from the viewer.

Nevertheless, AI is a powerful tool to generate surprising things. The results can be truly captivating when understood and put into context, i.e. mentally curated. AI products can be used as raw material for further collages or other physical and digital media works and thus be part of the creative process.

More importantly, the use of AI questions the quality of our digital image production - just as photography questioned painting and film questioned theatre. This "questioning" always took some time before something original and unique emerged instead of the clichés of the source medium. This means that we should not only use AI as a tool, but develop an awareness of the new things that may arise in it once it becomes clear why and how we want to use this technology.

That's why I'm very pleased that so many colleagues are interested in the phantasms of AI programs. Because this creates the right questions and thus possible anchor points in the image tsunami of the World Wide Web. Even before the AI wave, we were flooded with mood boards that were often contradictory and unclear because they parasitically simulated the creative results of complex art production but simply ignored their context. The main problem is the same with AI images - the mass use of high-quality image material at the expense of the conscious design process and thereby less access to the inner statics and construction of the images.

At the end of the hype about AI, there will be more and cheaper production of images! The quality of these image worlds is in our hands. At best, the use of AI will lead to a radicalization of design - just as classical modernism reacted to the image mush of eclecticism. I would like to see a punk revolution of images from unruly, controversial and emotional viewpoints, instead of an endless mush of AI mass production. The not so nice alternative would be for viewers to get used to the current image fast food. To counteract this, we must be able to look into the engine room of image production, put the artistic intention above the means and judge the result of the AI from the context of the content. This is the only way we can take responsibility as co-authors. For the production designers, visualization is a tool, not a result. The design shapes the image, not the image the design.

Excerpt from: "Production Design, Shaping the Invisible" AT, Chapter: "Sheep, Robots and Dreams, Artificial Intelligence and Design", Thomas Stammer, Berlin 2024