Two days out from a client meeting with only one design direction in hand is a stressful place to be. AI image generation now lets you fan a single concept out into a dozen directions in minutes. The catch shows up right after: run a batch of ten variations and it’s common to find only one or two actually worth showing anyone.
1. Fast and Good Are Not the Same Thing
Generate with AI and a single click produces dozens of results. The problem is that without a clear standard for choosing among them, reviewing that pile can eat up more time than sketching by hand ever did. Volume of variation doesn’t guarantee quality. A batch generated without direction just comes back around at the next meeting as “so what’s our concept, exactly?”
2. Structuring Prompts to Produce Real Variation
Producing variations efficiently means splitting a prompt into fixed elements and variable elements before you start. Change only one of form, material, or color at a time and hold everything else constant, and when you compare results afterward it becomes far clearer which change produced which impression. Midjourney’s own documentation breaks variation strength into discrete steps — understanding and using options like that turns a pile of random images into a set you can actually compare side by side.
Fixed elements
What stays untouched this round. Holding everything else constant is what makes comparison possible.
Variable elements
Pick one — form, material, or color — and change only that. Shift every parameter at once and comparison becomes impossible.
3. Decide the Selection Criteria Before You Generate
Skip deciding what you’re judging variations against before generation starts, and the more options pile up, the harder choosing gets. Whether this round should prioritize the user experience called for in the brief, manufacturing feasibility, or brand tone needs to be settled first — generation comes second.
Early concept exploration
Diversity of form comes first. Cast a wide net, then narrow toward a direction.
Right before presentation
Polish and consistency come first. Focus on refining the direction that’s already been chosen.
💡 Pro tip — AI-generated concepts often look sharp on screen and fall apart the moment they meet real materials or real manufacturing processes. Pairing this with a workflow that connects back to actual production tools helps filter out variations that are gorgeous but unbuildable. Screening for feasibility as early as the variation stage saves a round of rework later.
4. Build the Habit of Logging Variations
The more output accumulates, the more “wait, how did we make that one again?” becomes a recurring problem if nobody’s tracking which prompt and which settings produced which result. This bites hardest when a client asks to revisit a specific variation from an earlier round — without a way to reproduce it, the only option is regenerating from scratch until something similar turns up again.
Save the exact prompt text in a filename or a shared sheet
Log the generation settings too — seed value, variation strength — so results stay reproducible
Keep variations that didn’t get chosen instead of discarding them right away
💡 Pro tip — When direction shifts mid-project, a variation rejected in an earlier round sometimes turns out to fit the new direction perfectly. Keeping a log means pulling that candidate back out instead of generating everything from zero again.
5. Why Human Curation Still Matters
AI raising the speed of generation doesn’t remove the need for a designer’s judgment — if anything, the more options there are, the more it matters to have someone who can explain why this one got picked over the rest. A common mistake is dumping the full, unfiltered batch on a client and letting them pick blind. Choices with no direction behind them tend to slow decisions down rather than speed them up, and they blur whatever point of view the team had going in.
Hand off the raw batch
Share every variation and let the client pick. Decisions slow down and the team’s perspective never really shows.
Curate first
The designer narrows to a few against clear criteria, then presents the reasoning alongside them. Discussion happens on solid ground.
Closing thoughts
Did this round change only one variable — form, material, or color?
Did the team agree on selection priorities before generating?
Were the finalists checked back against the brief’s actual goals?
Were low-feasibility options filtered out early?
Was the final shortlist narrowed to three or fewer before sharing?
The ability to produce design concepts fast matters less than the habit of deciding, in advance, how those results will be judged. AI only makes the hand move faster — deciding what’s worth drawing is still, and will remain, the designer’s job.
Design Daily Life · Notes on design, daily