Gains:
- Ability to produce a wide variety of concepts and sketches in a short time with diffusion-based text-visual tools
- Ability to position artificial intelligence correctly and select ideas as a designer in divergent and convergent design stages
- Ability to consider artificial intelligence concepts as rough sketches and critically filter them in terms of form, function and manufacturability
The most exciting and risky phase of industrial design is concept production. "Concept" is an idea that shows the direction a product could be, the details of which have not yet been resolved; A "sketch" is a rough drawing that quickly visualizes this idea. A good designer doesn't stick to the first idea that comes to mind; It first generates as many and diverse ideas as possible (this is called divergent thinking), then eliminates them and focuses on a few strengths (convergent thinking). At this divergent stage, artificial intelligence can present in minutes the diversity of ideas that a designer alone produces in hours. But if this power is not used correctly, it will lead the designer to choose without thinking and to cliché. This unit explains how to use AI as an “idea accelerator” in concept generation and why you should retain selection judgment.
Divergent and convergent: At what stage does AI come in handy?
Think of the concept process like a funnel. There is a wide mouth at the top: as many ideas as possible. AI is amazing here; It produces 20 different form directions with a prompt and brings you combinations you would never have thought of. We narrow down to the bottom of the funnel: we screen ideas for function, ergonomics, manufacturability and branding. AI is weak at this contraction stage because it has no judgment about which idea will actually work. The rule of thumb here is that AI is the engine of the divergence phase and just a helper of the convergence phase; The choice belongs to the designer.
Here's how a step-by-step concept session works:
- Clarify the brief: What is the product, user, constraint, form language? (Explain this clearly to the AI.)
- Generate broadly: Ask for many variations (divergence) with different form languages.
- Cluster: Group outputs by form families (e.g. "soft organic", "sharp technical", "retro").
- Critique and select: Ask function/ergonomics/production questions for each family; Choose 2-3 directions (convergence).
- Develop by hand: Deepen selected aspects with your own sketch or CAD. AI output is the starting point, not the ending.
Tip: See what comes out of the AI as “stimulus” not “options”. The goal is not to find the best AI visual, but to catch the spark that will ignite your own design idea.
Setting up the prompt based on concept generation
A good visual prompt at the concept stage consists of several components: subject (what product), form language (minimal, organic, brutalist, retro...), material impression, angle and light, style (concept sketch or render) and number of variations. Keeping the same body silhouette but changing just one element (e.g. just the cover form) gives you a comparable set; Changing everything at once produces messy and incomparable results.
The following table shows how prompt emphasis changes for different concept intents:
concept intention
prompt emphasis
expected output
Broad exploration (divergence)
"10 very different form languages, same function"
Directions far from each other
within-family variation
"Same silhouette, only [X] is different"
Comparable set
Deepening a particular aspect
"This is close to the reference, 4 subtle variations"
Close alternatives
atmosphere/moodboard
"Not the product, but the feeling of material + light + texture"
inspiration board
three mini cases
Case 1 — Cliché trap. For a Bluetooth speaker, a designer simply types "modern speaker design" into AI and produces 30 images. 27 of the releases are close to the same cliché: cylindrical, fabric-covered, buttons on top. Because the vague prompt pushes the AI towards the average it sees most often. The designer rewrites the prompt as "non-geometric, stone/pebble inspired organic form, not textile but mineral surface, 8 very different directions" and takes 8 much more original directions. Lesson: diversity pushes the prompt; Ambiguity produces clichés.
Case 2 — Leaving the choice to AI. A team produces 40 AI concepts, votes for the “favourite” and takes it to the customer. However, the base form of the selected concept cannot physically accommodate the planned speaker driver; internal volume is insufficient. None of the 40 images considered interior architecture because the AI doesn't know that. The team starts over. Lesson: selection is based on function and physique, not just beauty; This judgment belongs to the designer.
Case 3 — Correct use. A designer predetermines 3 different form families for a thermos (cylindrical-minimal, faceted-crystalline, organic-drop). It asks the AI for 6 variations for each family, getting a total of 18 images. He uses these in a "direction selection" meeting with the client; customer chooses faceted family. The designer sketches this aspect by hand and figures out the number and depth of the facets in terms of production and grip. AI has accelerated discovery and humans have made the decisions.
Copiable prompt templates
LARGE DISCOVERY TEMPLATE "Concept sketches for [product]. Same basic function, but 10 VARIOUS form languages: minimal, organic, technical, retro, brutalist, toyish, luxury, industrial... Each one distinctly different. 3/4 angle, soft studio light, neutral background. Industrial design concept sketch style."
DOMESTIC VARIATION TEMPLATE"KEEP the silhouette and proportions of: [recipe/ref]. Produce 6 variations for [element to change: top cover / handle / base]. Keep everything else the same so it's comparable. Same angle, same light, same material impression."
DIRECTION DEPENING TEMPLATE"Stay close to the reference direction, but refine it: [ref/recipe]. Produce 4 subtle variations, the difference being only in proportion, radius (corner roundness) and surface separation. Radical change of form. Purpose: to mature the chosen direction."
CONCEPT CRITICISM TEMPLATE (for text AI) "Criticize the following concept idea from the perspective of an industrial designer: [describe the concept]. Ask questions on the following topics: function (does it really work), ergonomics, manufacturability (by what method is it produced, where is the problem), cost, originality. Not answers, but CRITICAL QUESTIONS that I need to answer."
Weak prompt / Strong prompt
WEAK PROMPT: "A modern desk lamp."
STRONG PROMPT:"Desktop LED work lamp concept sketches. Goal: minimalisthome office. Desired feel: lightweight, positionable with one hand, no cable visible. 8 MANY DIFFERENT directions for form language exploration: tilting arm, folding, swivel ring, counterbalanced, bending without joints... Material impression: matte metal + textile cable. 3/4 angle, softlight, neutral background. Concept sketch style; this is a It is discovery, measure, then it will be solved."
Poor prompting produces only one cliché; Powerful prompt variety and pushes design intent, giving you real directions to choose from.
Common mistakes
- Writing vague prompts and settling for clichés. Only clear instruction forces diversity.
- Changing everything at the same time and producing incomparable sets. Play one variable at a time.
- Making the choice based on beauty. Function, internal volume and production are also decision criteria.
- Thinking the AI concept is over. The output is an initial sketch; The actual design is your development.
- Leaving the hundreds of images you produce without any clustering. Selection cannot be made without clustering.
Caution: AI produces lots of attractive visuals; this abundance creates the illusion that "the decision has been made." Only you verify that an aspect is manufacturable, functional and original.
In summary
The concept phase consists of two movements: divergence (many ideas) and convergence (elimination). AI is the powerful engine of divergence: it yields a large and diverse pool of ideas in minutes. But it is poor at convergence, that is, at deciding which idea will actually work; This judgment belongs to the designer. Force variety with clear and structured prompting, cluster outputs into form families, choose based on function/ergonomics/manufacturing rather than beauty, and develop your chosen direction by hand or with CAD. The AI concept is a spark, not the finished design.
Application task
Choose an item (for example, a water bottle). First, define 3 different form families in your mind. Produce 6 variations for each family with the "broad exploration" template. Cluster the resulting 18 images by families and extract function/ergonomics/production questions with the “Concept critique” template for each family. Choose the 2 strongest aspects, write down why you chose them (other than beauty), and refine each of these two aspects with a small sketch by hand.
checklist
- [ ] I wrote a clear and diverse prompt for Divergence.
- [ ] I produced a comparable set by moving one variable at a time.
- [ ] I clustered the outputs into form families.
- [ ] I evaluated the choice in terms of function, ergonomics and manufacturability, not just beauty.
- [ ] I didn't leave the direction I chose with the AI output, I developed it by hand/CAD.
- [ ] In terms of originality, I questioned whether the output was a cliché or an imitation.