AI as designer's toolset
Artificial intelligence is an increasingly prevalent topic. As it continues to evolve, its role in various industries, including product design, cannot be ignored. But after getting to know these new technologies, we have one big conclusion: rather than shunning AI out of fear or spite, designers must engage with it to harness its potential.
The fear of AI replacing human workforce
The rise of AI has sparked discussions about its capacity to replace human jobs. However, we believe that the reality is different. AI cannot replace the nuanced understanding and creative insight that human designers bring. Instead, it serves as an aid, helping to enhance the workflow.
AI is still just a tool at the end of the day. While it can do spectacular things, turning its outputs into meaningful ideas isn't as easy as pushing a button. To design a useful and appealing product, experts are needed who understand what it makes to match user preferences, pain points, and other kinds of expectations. There are so many details that needs attention during a design process that it simply cannot be done by one platform.
Of course, one can say that while this may be true in 2024, we cannot truly know what future will bring: maybe in a few years’ time AI will be capable of a more holistic thinking, that can prove to be a real danger to human workforce. We cannot make predictions, but consciously keeping up with the changing of times seems like a generally useful approach when dealing with new technologies. The longer we are involved in a rapidly changing landscape, the better.
As for now, we can view the introduction of AI in design as akin to the advent of 3D printing or Photoshop. It opens new horizons, transforms certain processes, and accelerates others. Yet, it doesn't diminish the importance of human creativity and thoughtful control.
AI generated illustrations for storyboarding
Enhancing the workflow
Using generative AI for visualizing concepts requires understanding its limitations. While it can produce high-quality, polished results, it's not always obedient to our exact desires. The details may be incorrect, as they are randomly generated. This can trick users into thinking it cannot be tweaked. However, it should be tweaked, as it can generate something infinite times. This disobedience should be embraced instead, with constant jumping between AI idea generation and manual sketching.
According to our experience, image generating AI is most useful during the initial ideation process, generating variations of shape ideas and atmospheric in-situ illustrations. Ideally, a designer will see a promising gesture, a small detail in a generated image that can spark further ideation on their part, but this won’t be enough to stand on its own. It doesn’t matter how ready the results seem at first glance; they always need tweaking and refinement at least. In most of the cases they are not useful at all, but as they can be generated in great numbers easily in the background, by chance the user will soon notice something that has potential.
Other phases of the design process are usually way too detailed for the AI to help. It can browse through large amounts of research data to show a summarized picture, but it is way too random for a useful input during the developing and refining process.
As many users have already noticed, there is only so much detail that an image generating platform can accepts as input, so after a certain number of words they randomly select & mix parts of a lengthy prompt. Generative AI has its unique way of interpreting information, which may not always align with human descriptions. A potential bypass is to generate more images using simpler prompts and manually morphing the results before jumping back to an AI platform for further work.
Finding the right platform is essential. Many options are available, each with a different toolset. As the famous “jam experiment” study showed, too many choices can overwhelm the consumers. This is true for output from generative AI as well as for the number of generative AI platforms available. So, which one works the best?
Moodboard to feed the AI platform
As far as we experienced, most image-to-image platforms are unsuitable for industrial design purposes since they are designed for creating human characters and landscapes (mainly in photoreal or anime style). As a result, they may not provide the necessary tools and features for industrial design applications. We should look for platforms that embrace the ‘hallucinating’ quality of AI, so that we simply have more randomized output to work with later. One can say that this is basically an AI-aided brainstorming, not a final solution for producing complete design structures. Platforms like Hypersketch, Vizcom or Firefly excel in facilitating this back-and-forth play between AI-generated concepts and traditional refinement.
Multiple results where selection is needed
Achieving the correct details of the subject, the exact environment, lighting, color, mood, and composition as described in the brief still depends on experts. AI only makes the whole process faster or less mundane, but at this moment the results it gives won’t make sense without the careful eye of a designer.