Integrating AI into my industrial design workflow…
Okay, so this one has been in the works for a while now, with new AI tools seemingly popping up left, right, and center. Right now, there seems to be a new tool, program, or development each week, and with so much noise, I’ve spent a little time getting to grips with some of them. I’ve been keen to find out how they work best, and I’ve looked into different ways I can implement them into my design workflow.
As a disclaimer, these are just a collection of my thoughts and views. I’m sure there are plenty of other tools out there, each offering something different or operating in different ways.
My focus was to try to understand these tools better, studying various ways they could work for me and their most suitable application.
To begin with, I found it best to put together a list of points and questions where I could see AI potentially adding the most value.
.
So where did I start?
For me, time is incredibly important, and I never seem to have enough of it. I was interested in finding a way where I could leverage different AI tools to utilize them as some form of design assistant. The key areas I was most interested in exploring were:
Integration
How do these tools work best and what part of the design process are they most suited to?
Efficiency
How can they save me time?
Productivity
How can they increase my level of work output whilst retaining a consistent level of quality?
After studying these areas and briefly researching some of the top go to tools other professionals have been using I started to ask myself the following:
What are the pain points for me in my process?
Are there any tasks which I seem to find take me longer than I originally intended?
My focus was not around how AI can replace some of my tools. It was more about freeing up time within different parts of the design project to allow me to add increased value.
I’ve found the best way to really get under the skin of these tools and to understand their capabilities is to have a play and experiment. A quick exercise enabled me to quickly identify some useful features and different advantages of some of the tools.
Sure, some struggled to interpret my original design direction in some areas but excelled in others. I’ll be honest, on initial first impression, I was finding some of the tools more frustrating than anything else. The results were far too variable, and the lack of control was proving a hindrance and time-consuming. However, once I began to understand the translation of various image and text content better, I was then able to apply design direction into different areas and push the AI engine.
At this point, what I began to realize is that AI is a great enabler, but the tool is simply just that—an enabler.
Like Photoshop, Illustrator, or Sketchbook, just because you have access to these tools does not make you a great designer.
It’s the application of the tool in the right part of the process which will make the biggest impact.
As designers, it’s our job to ask questions, critically evaluate, observe, and validate ideas.
For me, a natural step was to critically observe and evaluate AI output. I began to view AI as more of a design assistant in a way where I was critiquing output and using the information it was throwing out as references.
Thinking of the output more as design suggestions is a better way of framing the tool for me.
It’s worth noting in all of this that AI will not throw up anything that is novel and unique as the content is culminated through a use of different sources. If you have a solid idea and your design concept explores some innovative features, the results AI generates will likely fall short of your expectations.
This can be frustrating from a designer's perspective, and from my experience, it’s far better to enter the AI world with an open mind and use the tools suggestively in an exploratory way.
This leads me onto another point…
These tools still require a level of design input and creative steer. It’s very easy to find concepts and design images with gaping holes and glaringly obvious practical flaws. The key here is to be specific and concise in your input. The visual content generated is very good and can be super high quality, but that doesn’t mean to say your design is good.
Design is about validation and consideration, something AI all too often misses.
On that note, I imagine it would be quite easy to fall into a trap where AI became something more of a reliance in your toolkit. This is worrying from a design perspective, are we likely to see a ton of unconsidered, unresolved products churned out by the AI engine….
That’s another topic entirely…
Understanding the capabilities and the power of the tools but also acknowledging its limitations will help you to achieve the best results and also help to save you time in the long run.
This led me to think…
Typically with any design project there are four stages: Discover, Define, Develop Deliver
Although project’s don’t always operate start to finish in this way the framework is a good tool to visualise the process. I can see AI having the most value feeding into 3 of these areas and here’s a list of tools which I’ve attributed to each.
Discover
Moodboards, styling references, competitor & trend analysis and market research
Mooed AI
Midjourney Explore
Midjourney
Keplo
Chat GPT
Define
Concept generation and concept visuals
Krea
Midjourney
Vizcom
Hypersketch (Ipad)
Chat GPT
Delivery
Product visuals, presentation material, marketing visuals
Midjourney
Photoshop AI
Vizcom
Krea
Chat GPT
Demo Time…
Here’s a fairly quick use case example showcasing the process I would go through typically at the beginning of a design project. Again, this approach may not be suitable for every type of project, but let’s explore for now.
Allocating a section of time early on like this can prove a worthwhile and a valuable exercise. I’ve found that a quick experimental design exercise can help to spark some new ideas, thoughts, or suggestions in terms of concept direction and creative steer for a design project.
As a rule of thumb, I’ve found it best to use a hybrid approach switching from AI to digital sketching, image manipulation, and back again.
I’ve found this approach useful in providing the most consistent and accurate results.
Again, I’m not saying this is the only way to use the tools or in fact the best way but for me, I’ve found the results to be somewhat consistent, and I can certainly see value in its integration.
Lastly, it’s very easy to get caught up down a rabbit hole with AI.
Knowing when to stop and pull away from the tool is just as important as knowing when to pick the tool up and begin with it.
It’s easy to fall into a trap where you’re constantly inputting information and not getting valuable output which can be both incredibly frustrating and time-consuming.
Understanding the right time to collate your findings, pull away and feed the information back into your process away from AI is incredibly important. This skill becomes far easier over time and use, and generally, you’ll get a good feel for when this point is with greater exposure.
Case Study: Screwdriver Handles
Midjourney > Mooedai > Sketchbook > Krea > Midjourney > ChatGPT > Midjourney
Project Objective: Re-design the handle of a screwdriver, aiming to create a contemporary, sleek, dynamic, and ergonomic design. This project intends to elevate the user experience while maintaining functionality and practicality.
1. Moodboard Generation (Midjourney) - Rapid explore
Prompt used - Create a moodboard of modern screwdriver hand tools in a black and orange colour scheme, white background, contemporary, sleek, dynamic and ergonomic design.
Note: Deconstruct the brief, find keywords within the project description and user requirements. Keep the prompt simple and concise at this early stage.
2. Moodboard Generation (Mooed.ai) - Refined image explore
3. Moodboard Generation (Midjourney) - Combined focus explore (Varied image weights, — iw.5, —iw1, — iw2)
Generating a hybrid moodboard, uploading the mooed.ai moodboard to the Midjourney server.
<Insert image link> Create a moodboard of modern screwdriver handtools in a black and orange colour scheme, white background, contemporary, sleek, dynamic and ergonomic design.
4. Quick loose concept explore - Underlay ref
A very quick sketch exercise taking inspiration from moodboards generated in midjourney and placed alongside a reference image to understand proportions. This sketch is very rough and took around a minute and a half to put together. My areas of focus was the overall silhouette and form identifiers. The idea behind the sketch is to produce something quickly AI will be able to interpret and work with.
5. Sketch transfer into Krea - Rapid visualisation output
Prompt used - A modern screwdriver hand tool in a black and orange colour scheme, minimal and ergonomic design.
Note: Prompt description has been simplified, AI strength is approximately 45% (AI strength increases or decreases AI influence)
6. Upload Krea output to Midjourney and use the describe image tool.
Use the describe tool within Midjourney to interpret the image.
This will instruct Midjourney to pull together 4 descriptions based on the image input. From there I’ll go through the descriptions to find the most suitable which fits the design in the image..
Note: I chose description 4, nice, simple and concise with enough room for further adjustment if necessary.
7. Check your Midjourney settings.
This is fairly simple to do. Type /settings into the message bar and your settings options will appear in a loaded response, simply toggle these settings to adjust, depending on your preference.
I like to keep stylize low at this stage with a low variation mode. My aim is to create as many varied consistent results as possible.
8. Link your descriptor to the original keywords in the moodboard to enhance your prompt with ChatGPT.
Note: I’ve used ChatGPT to merge my prompt descriptor in midjourney and additional text used to generate the moodboards at the beginning of the design excercise.
9. Create your design concepts.
<Insert image link> .Pov front view of an orange and black screwdriver with a contemporary, sleek, dynamic, and ergonomic design, isolated on a white background, ultra realistic photography --iw 2 --s 50 --style raw
Note: The image weight parameter (—iw X) adjusts the importance of the image against the text (.5 is a low image weight and 2 is high). I like to use 2 to retain the closest visual link.
Typically for this kind of exercise that’s probably as far as I’ll go in AI for now…
Note: I wouldn’t present these concepts to a client straight off the bat. These concepts have to be studied, pulled apart and modified where necessary. There are certain details which don’t make sense and others which are quite nice. I’m not entirely sure what’s going on with the screwdriver tips for example.
As a fairly quick excercise there are some form identifiers there which link really nice to the moodboards generated at the beginning of the design exercise. Typically what I’d now do is take these design concepts into a different software package and work them through.
This is just one use case example and there are others I’ve been experimenting with recently.
If you’re looking to accelerate your concept generation work, please get in touch, fill in the contact form and let’s talk design!
