The cost of waiting on AI is partly the practice your team never gets. A competitor can buy the same software tomorrow. Learning which results deserve approval and why takes practice; six months of it gives a team judgments it can use on the next assignment. Canva’s new image generator offers a concrete example of what that investment looks like, even for businesses with no intention of building their own AI.
Canva released Craft on October 22. The headline is another company entering image generation. The more interesting detail is how Canva brought its designers into the work.
In brief
Canva says its designers and researchers assessed more than 25,000 image pairs while evaluating Craft. Your business can begin at its own scale with a recurring assignment, an experienced reviewer, and a way to turn repeated corrections into something the whole team understands. Start there before counting AI subscriptions as capability.
The person who knows why it is wrong
In its launch announcement, Canva says professional designers helped improve instruction following, typography accuracy, and visual quality. For internal evaluations, they worked with researchers to compare more than 25,000 pairs of images against leading alternatives. When automated evaluations disagreed with designers, Canva says it sided with the designer.
That last decision matters. A result can score well on a general test and still be wrong for the job.
A marketing director knows when a campaign looks polished but says nothing. An account manager recognizes a customer brief that omits the one promise likely to cause trouble. Neither judgment is captured by asking whether the AI produced a complete-looking answer.
Ask that account manager what went missing. The answer might be a sentence everyone else skimmed past, and it belongs in the brief for the next assignment while people still remember why it caused trouble.
Canva’s process offers a product-development example; marketing gains for your team would need their own evidence. The comparison figures are company-reported, and they do not establish time savings for your team. They do show a company making professional judgment part of product development instead of treating it as cleanup after generation.
Access arrives faster than judgment
The Verge reports that Craft produces images at up to 2K resolution in more than 30 aspect ratios. Canva says it can create text in 55 languages. Those capabilities become available to many businesses at roughly the same time.
Knowing what to ask for, and recognizing what should never reach a customer, develops less evenly.
Imagine two regional businesses using the same AI to prepare campaign materials. This example is hypothetical.
One team generates options, picks the nicest-looking version, and fixes problems as they appear. Its strongest designer quietly repairs the awkward work before publication. Everyone is grateful. Nobody else learns much.
The other team asks that designer to explain a few decisions during an existing campaign review. Why did this layout fail? Which wording changed the offer? Why does the technically correct version still feel wrong for the audience?
After several campaigns, newer colleagues can recognize some of those problems themselves. They give better direction earlier and send fewer weak options to the reviewer.
The first team’s designer is still the bottleneck. Freeing that person’s time requires sharing the judgment behind those repairs with the team.
Give the expert time to teach
The owner should be the person responsible for the quality of that recurring assignment, with time allocated for review. For a campaign, that might be the marketing lead. Technology staff can help with approved tools and data handling; they cannot decide what a convincing offer looks like for your customer.
Start with one familiar deliverable. Use material the team is authorized to put into the chosen system. Keep normal approval requirements in place.
At the next review, compare two plausible versions against the same brief. Ask the experienced reviewer to choose and explain the difference. If the answer is “This is more on-brand,” ask which details make it so. “The headline promises a discount the offer doesn’t include” gives everyone something they can check next time.
Keep the explanation beside the example. Bring it back when the next assignment arrives. See whether a colleague can apply the lesson without the expert reconstructing the conversation.
Keep doing this during ordinary reviews with the tools you already use. The examples the work provides can help the expert gradually stop catching the same error for everyone else.
Count what stops coming back
Generation speed is easy to notice. Better review habits take longer to see.
For a small trial, record the time between first draft and approval, including the reviewer’s time. Note which mistakes recur. A faster first draft that needs more repair may leave the expert worse off.
The useful change is a recurring mistake that stops returning, or a colleague who can explain why a draft needs another pass before sending it upstairs. Those are signs that the team is learning how to judge the options it produces.
That is the part of human-AI collaboration in practice worth making time for: colleagues learning enough from each review to handle more of the next assignment themselves.
Put the next campaign review on the calendar with enough time for the designer to explain a rejected version. Save that example. A few months later, ask whether anyone still needs the explanation.
Research and structure: Mai. Editorial direction and voice: John Lipe.