As AI-powered image creation tools are emerging and advancing faster than one can follow, a question comes to mind: could any of these tools serve for concept development for research and rapid prototyping?
The role of visual prompts in design research & prototyping
Visual prompts are extremely helpful for design research. They enable us to move beyond questions & answers, towards a more natural, in-depth conversation with participants – leading onto insight in people’s perceptions, attitudes and needs. Especially in cross-cultural research, working with sketches or printed illustrations can help bridge linguistic and socio-cultural barriers. However – these illustrations need to be simple and culturally relevant. A drawing of a farmer in Africa will not be understood – if their clothes and appearance are based on western references. Similarly, photographic or complex details can become a distraction for our conversations and fail their purpose.
When prototyping for service design, I often use storyboards to communicate a high-level service concept and to gather feedback before developing particular touchpoints in more detail. So how can AI help?
A day of experiments with various AI tools (openart.ai, playground.com, nightcafe, craiyon among others) was quite revealing. To be consistent, I prompted all the AIs to draw an illustration of “An Indonesian man placing a plastic bag with trash on a sandy beach fringed by palm trees” (with some tweaks and additions about flat illustration, simple shapes etc, when necessary). Why? Because I wanted to visualise a behaviour that I have observed in some remote parts of Indonesia, for example on Hatta island, in the Banda islands. I saw a few locals clean the beach from plastic debris, then put everything in a plastic bag…and then leave the trash bag by the sea for the “tide” to pick it up. The following day, the “problem” was gone. This behaviour is intriguing in its own right, suggesting some level of awareness that plastics need to be collected and that the beach needs to be clean (assuming for tourists to come and enjoy their stay), but failing to perform waste disposal and opting for waste displacement instead (i.e. putting it back into the ocean). Anyway, I could go on a separate blog about the drivers and barriers for this behaviour – but for now let’s just focus on this as an example worth visualising.
So my test for AI was to see if it can represent the observed behaviour of placing a trash bag on the beach in one simple illustration. So here is what I found out:
1. AI models desperately need ethno-cultural and geographical references
Try prompting the illustration of an Indonesian man (or various other nationalities) and you are in for a surprise. In some tools he might look African, in others a bit more Japanese…so hiding the face under a hat probably delivers the best result (even though a man on the beach would not wear this type of hat – it is mostly used for working in rice paddies). But this is to be expected. After all, most freely available image and illustration libraries tend to fall short in the representation of non-white or non-western communities. Graphic design suffers from stereotypes and AI has inherited this W.E.I.R.D point of view. Since my design research spans across borders and is largely focused on the Global South, I need to be able to “feed” AI with the right cultural references. For example, I would like to be able to upload a few photographs as references of environmental context, clothing, housing, facial characteristics and so on. And most importantly, I need AI to be loyal to those cultural references. At the moment, the outcome is quite unpredictable and somewhat chance-dependent.
2. Photorealism works against “simple”
We are in a phase of infatuation with AI visual creativity: superb photorealism, mixing and matching styles, imagining unreal concepts that “look real”. It’s impressive what creatives can generate with AI tools nowdays, using creative prompt engineering and a combination of platforms.
But in my design research practice, “less is more”. I don’t want research participants to be distracted by beautiful textures and illustration details.
I don’t need art, I need design. A key insight across AI tools, is how difficult it was for me to find the one style, the one setting that would deliver a “simple illustration”. Or at least some sorts of sketch that would not look photographic, but neither child-like. And being able to find this option, before I ran out of free credits on the AI platforms that I have been testing.
3. Design iteration is key and AI still needs to learn “subtraction”
Design is an iterative process – this also applies to conceptual illustrations and visuals for research purposes. Iterating might mean modifying particular elements, adding more detail or removing distracting objects. If I visualise a given behaviour for research purposes, I want the audience to focus on the action itself. An iterative design process to fine tune details is very difficult with AI tools – even if you increase the “prompt compliance”. In principle, you should be able to select particular elements and request their removal or replacement. For example, I highlighted an area on the sand and asked AI to remove the trash from the beach (because the trash is supposed to have been collected in the bag). This resulted in a blue smudge..then in a pile of sand… but no consistency. Perhaps, this is an unfortunate anthropomorphic feature, but AI seems to be much better at adding more visual details rather than subtracting them.
So when I got the illustration below, I smiled. It almost felt like I am working with a talented but very junior designer or intern – who just can’t help it, but keeps adding random decorative elements. Birds hanging from the sky?
4. There is no AI tool yet for storyboarding for prototyping purposes
I often work work with storyboards for service design. They are great as prototypes for testing new services and co-creating them with communities. But storyboard prototypes need to be high level, simple and consistent in their representations. A prototype has several iterations, so these storyboards need to be imperfect, so that people participating in co-creation workshops can critique, modify them and enrich them with new ideas. In our example, of the islander in Indonesia leaving a bag of trash for the tide to carry away, we would need to represent various “solutions” as part of a story. What would be a potential intervention that would encourage this person to change their behaviour? In order to visualise and test ideas, we would need to re-use the same character(s) across different situations and settings. But we also need simplicity, to get the key ideas across in order to evaluate a storyboard as a prototype with local communities.
I have yet to identify an AI model or tool that enables re-usability of elements for storyboarding and simplicity in representation.
In conclusion, in order to serve design research across borders and cultures, new AI models and different approaches to AI-enabled design tools need to be developed. Although I believe it is still early days, I will continue to experiment with AI to create visuals for research purposes.
Have you tried any AI tools to create high level prototypes?
What was your experience like?
Thoughts and suggestions are most welcome!
This article was originally published on LinkedIn, on 14/05/2024
Leave a Reply