I Ran Canva AI 2.0's Brand Intelligence Against the Monthly Brand-Refresh I Bill Studios For. Here's What Held in the Research Preview and What I Reopened Figma For.
I ran Canva AI 2.0's Brand Intelligence against the monthly brand-refresh I bill studios for. What held in the research preview, what I redid in Figma.

Every time a studio client signs off on a brand tweak, a new accent, a different display face, a tightened logo lockup, I inherit the same unglamorous job: re-skin every live template they own. Social kit, pitch-deck master, one-pager, ad set. It is roughly four hours of a mid designer's time per client, it recurs almost monthly, and it is the single line item clients hate paying for. Canva AI 2.0's Brand Intelligence claims to do it "in seconds." So I ran it against the actual job.
The job Brand Intelligence says it kills
The recurring deliverable nobody photographs for a portfolio is the brand-refresh re-skin. A client ships a v2 of their identity. Not a rebrand, a refinement. The accent moves from a flat teal to a slightly warmer cyan, the display face changes from a grotesque to a tighter neo-grotesque, the logo gets a new clear-space rule. Nothing conceptually hard. But it touches every asset that already exists, and at a studio running several active clients that is a few hundred layouts a quarter.
Canva's pitch for Canva AI 2.0, announced April 15, 2026 and currently a research preview with general availability rolling out over the following weeks, names this job directly. Brand Intelligence, in Canva's words, lets you "instantly update existing work by asking Canva AI to apply your latest brand, turning hours of manual updates into a single step done in seconds." That is not a vague productivity claim. It is a specific bet on the exact task I bill for. Worth testing properly.
What Canva AI 2.0 actually is under the marketing
Strip the keynote and there are two architectural changes that matter to a brand-systems workflow.
The first is what Canva calls the Canva Design Model, which it describes as a foundation model built to understand "the structure, hierarchy, and complexity of real-world design." The practical consequence is the second change: layered object intelligence. Output is assembled "from individual, editable objects rather than generating a static or locked flat image." Anyone who has worked with image models knows the failure that kills: you love 85 percent of a generation and cannot touch the other 15 without regenerating the whole thing into something different. Canva's predecessor step toward this shipped in March 2026 as Magic Layers; 2.0 is the architecture-level version.
Brand Intelligence sits on top of that. It reads your Brand Kit, fonts, colors, the style ramp, and applies it natively rather than asking you to restate it in every prompt. NAV43, one of a small set of Canva Agency Partners, framed it bluntly in their Canva Create 2026 review: for agencies running multiple client brands at once, native Brand Kit application "fundamentally changes the workflow." They were also honest that the 2025 AI features "did not land," which is the right amount of skepticism to bring into a preview test.
The test: one brand change, fourteen assets, research preview
I did not run this on a live client account. It is a research preview, so I rebuilt a representative brand world I control: a cool near-monochrome system, deep neutral ground, off-white type, one warm cyan accent, a neo-grotesque display face over a humanist text face. I assembled fourteen real asset types from that kit inside Canva: four social formats, a six-slide pitch master, a one-pager, an ad set in two ratios, an email header, and a simple data slide with a bar chart.
Then I made the brand change a client would actually request. The accent shifted one step warmer. The display face swapped to a tighter cut. The logo lockup gained a fixed clear-space frame. I wired the Notion connector so the live design-token table fed the model context instead of me pasting hex values, the connector list at preview covers Slack, Gmail, Google Drive, Google Calendar, Notion, Zoom, HubSpot, Microsoft, Atlassian and Linear. The instruction I gave Brand Intelligence was deliberately the instruction a non-designer client would give:
Apply our updated Brand Kit to every asset in this project.
New accent, new display typeface, new logo clear-space rule
are in the connected Notion token table. Keep layout and copy
unchanged. Update color, type ramp, and logo placement only.That is the honest version of the claim under test. Not "design me something," but "carry my v2 across what already exists, do not get creative."
What held
The flat, Canva-native layouts re-skinned cleanly. The four social formats and the one-pager picked up the new accent and the type ramp correctly on the first pass, including the secondary muted-gray steps, not just the headline color. This is the part the press releases get right. When the source asset was already built from native layered objects, token propagation worked the way you want a design system to work.
The single most valuable behavior was not the bulk pass. It was object-level correction without regeneration. The pitch master's cover came back with the new face but loose tracking, the optical kerning on a 64pt display line was visibly off. In the old image-model world that means regenerate and lose everything else. Here I selected the headline, tightened tracking, and nothing else moved. That is the actual paradigm shift for production work, and it is understated in the announcement. The regenerate-from-scratch tax is what made AI creative tools unusable for client work. Removing it changes the economics more than any single generation quality bump.
The Notion connector pulling the token table also held. Feeding a live source of truth instead of pasted values is the correct pattern for brand systems, it mirrors the connector-as-token-source approach designers are already adopting in other tools, and it meant the accent value was right because it read from the table, not from my memory of the table.
What I reopened Figma for
Three things broke, and the pattern in what broke is the useful finding.
The logo lockup did not recolor. It had been placed as a flattened export, and Brand Intelligence treated it as image content, not as a recolorable object. Correct behavior, arguably, but it means the "apply my latest brand" promise quietly excludes any artwork that entered Canva as a flat asset, which in real client handoffs is most logos. I rebuilt the lockup as native objects in Figma and reimported. Roughly twenty minutes.
The data slide kept the old accent in the chart series. The layout recolored, the bars did not. Templated chart components appear to sit outside the object graph Brand Intelligence reaches. I fixed the series by hand.
The display-face substitution was metrically close but not optically managed. On body sizes it was fine. On the two large display lines the new cut needed manual tracking and one forced line break the model did not make. This is the predictable seam: the model swaps the typeface, it does not art-direct the typeface. That distinction is the whole job.
None of these are damning. They are exactly the 15 percent you would expect a model to miss, and notably it is the 15 percent that needs a designer's eye anyway. The honest limit on top of all of it: this is a research preview. Persistent memory did not carry my brand context cleanly into a fresh session, so I kept the Brand Kit as the source of truth rather than trusting the model's memory. I would not put this on a billable client deadline until general availability and a few weeks of stability behind it.
The time ladder
My current re-skin workflow for a fourteen-asset set is a Canva-template and Figma round trip. Measured honestly against my own timesheet, it runs three and a half to four hours of mid-level designer time. At my studio's blended internal rate of 95 dollars an hour that is roughly 360 to 380 dollars of labor per refresh, and the refresh recurs close to monthly per active client.
The same set through the Canva AI 2.0 preview: the bulk re-skin pass completed in minutes. My hands-on correction time, the logo rebuild, the chart series, the two display lines, the kerning, came to about 50 minutes all in. Call it one hour against four. On this specific job that is a 70 to 75 percent time cut.
The number that matters is not the 75 percent saved. It is what the remaining 25 percent is made of. It is type optics, flattened-artwork edge cases, and the judgment calls. That work was never the cost problem. The cost problem was the tedium tier underneath it, and that tier is what collapsed. Pricing for the preview is not finalized, so I am not putting a tool cost on the ladder yet, the labor delta is the real line anyway.
The takeaway in design terms
Brand Intelligence does not replace the brand designer. It deletes the part of brand-system maintenance that was always beneath the designer's skill and always on the invoice. For a studio, that reframes the refresh line item: stop billing four hours of propagation and start billing one hour of art direction plus a higher-value strategic pass you previously had no time for. The tool does not change what good looks like. It changes how much you pay to keep it consistent. That is the part of brand work that should have been automated a decade ago, and is the kind of pipeline shift I build into client engagements at DVNC.studio.
If you want the prompt scaffold and the Brand Kit token structure I used to make the connector hand-off behave, it is in the AI poster and campaign prompt pack.
Is Canva AI 2.0 generally available yet?
No. As of mid-May 2026 it is a research preview, with Canva saying general availability rolls out over the following weeks. Treat preview results as directional, not contractual. Do not schedule client deadlines around it yet.
Does Brand Intelligence actually read my Brand Kit automatically?
Yes for native, layered Canva designs. It applied fonts, the color ramp and style without re-prompting. It does not recolor flattened artwork such as logo lockups imported as PNG or templated chart series that sit outside the object graph.
What is the real win, the bulk pass or something else?
The object-level correction without regeneration. Fixing one headline's tracking without disturbing the rest of the design removes the regenerate-from-scratch tax that made AI creative tools unusable for client production.
Will this remove the designer from brand maintenance?
No. It removes the propagation tedium. Type optics and edge-case artwork still need a designer's hands, and the art-direction judgment on top still needs a designer's eye. That is the 25 percent it cannot do, and it is the 25 percent worth paying for.
May 19, 2026
More from Design

Higgsfield's Supercomputer Ran My Client Brand Film End-to-End. Every Skill the Hermes Agent Took Over, Where It Drifted Off the Brand World, and the Credit Ladder Before I Pulled the Edit Back
A real billable brand-film brief run through Higgsfield's Supercomputer: what the agent handled, where it drifted, the credit ladder, when to pull back.
May 18, 2026
Canva AI 2.0 Made Brand Intelligence Agentic. I Pushed a Real Client Brand Kit Through 60 Assets and Counted Every Credit It Burned Before Drift Set In.
Canva AI 2.0 Brand Intelligence tested on a real client kit vs a Figma plus Midjourney pipeline: where it drifts and the credit cost per on-brand asset.
May 17, 2026
Canva AI 2.0 Builds Layered, Editable Designs Now. The Billable Studio Work It Quietly Eats, and the Two Slices It Can't Touch.
Canva AI 2.0's layered design model for a studio principal: the billable deliverables it concedes, the two it structurally can't, and the export wall.
May 17, 2026