Canva Just Shipped Its Own AI Image Model — What Craft Actually Changes for Small Business Marketing
By UVIDSTUDIO · October 29, 2026 · 7 min read · AI & Automation
On October 22, Canva launched Craft, its first in-house AI image generation model, live now in Dream Lab for every account and rolling out through a public API. Here's what it actually does, where it's strong and weak compared to Midjourney or DALL-E, and what that means for the ad creative, flyers, and social graphics your business already makes in Canva.
If your business already makes its social graphics, flyers, or ad creative inside Canva, the AI image generator sitting one tab over in that same account just changed. On October 22, 2026, Canva launched Craft — its first image-generation model built in-house rather than licensed from OpenAI, Google, or another outside lab — and shipped it live, immediately, into every account's Dream Lab tab.
What happened
Canva released Craft on October 22, 2026, and TechCrunch reported it as the company's first self-built image model, available now to every Canva user inside Dream Lab, with a public API rolling out so developers can build Craft into their own marketing platforms, e-commerce sites, and print tools. Canva also plans an MCP connector so third-party chatbots can call the model directly. The specs: images up to 2K resolution across more than 30 aspect ratios, built around a planning stage that works out composition before it renders — closer to how a layout tool thinks than how a typical photoreal image generator works, per BetaNews and Let's Data Science.
The detail that matters most for how the model behaves: Canva trained Craft on human-made design templates, not photographs, according to TechCrunch's reporting. That is a deliberate bet on what the model is for — posters, social cards, icons, and layout-driven graphics — rather than competing head-on with photoreal generators. Craft debuted fourth on DesignArena's crowdsourced leaderboard, which Canva says puts its research team among the top labs for image generation; cheaper Craft-fast and Craft-ultrafast variants are planned, according to AI Weekly. One outlet frames the move plainly: owning the model cuts how much Canva pays third-party AI labs every time a user clicks "generate" — a cost Canva has absorbed for years running other companies' models inside Magic Media.
Oct. 22 the day Canva shipped Craft, its first in-house AI image model, into every account's Dream Lab tab
Why it matters for small businesses
Most small businesses and nonprofits already treat Canva as their default design tool — the place the Tuesday social post, the event flyer, and the one-pager get made, as we've covered in our Canva Pro setup guide. Craft lands directly inside that habit rather than asking anyone to adopt a new tool, which is exactly why it is worth a few minutes of attention even if you never touch the API. Because it was trained on design templates rather than photography, Craft is likely to be stronger at the design-style graphics Canva users actually generate daily — posters, cards, icon sets, layout elements — than at photoreal product or lifestyle shots, which is a meaningfully different strength profile than Midjourney, the tool most businesses currently reach for when they need AI imagery with real visual range.
The bigger question is one we've raised with every AI image and logo tool on this blog: ownership. Craft-generated images, like Midjourney's or Looka's, come out of a model trained on a large shared pool of material and generated through prompts any other Canva account can also run — the same structural issue the U.S. Trademark Office has routinely used to reject marks built from non-exclusive, template-assembled artwork. An AI-native image generator living inside your existing design tool makes it easier than ever to drop a Craft image into an ad, a flyer, or — worse — a logo without stopping to ask whether you actually own what you just made. That question gets more urgent, not less, the more convenient the tool becomes.
What to do this week
None of this requires switching tools or panicking about your current Canva workflow — it just means using the new option deliberately instead of by default.
- Try Craft inside Dream Lab on a low-stakes graphic first — a social post or an internal deck slide — to see how it handles your brand's actual layout needs before you lean on it for client-facing work.
- Keep Craft, and any AI generator, out of your logo and core brand marks. The trademark risk that already applies to Canva-assembled logos and AI logo makers like Looka applies just as directly to a Craft-generated mark — origination still has to come from a human design process if you intend to register it.
- Check Canva's usage terms for Craft specifically before shipping generated images in paid ad creative, the same diligence we'd recommend before running any AI-generated asset through a campaign — terms and commercial-use conditions on new generative features tend to firm up in the weeks after launch, not at launch itself.
- If you build software or run an e-commerce catalog, watch the public API rollout — a design-trained image model reachable by API is a genuinely different integration than calling a general-purpose image model, and it may be worth evaluating once access widens beyond the initial rollout.
Craft is a real product move, not a marketing stunt — Canva now controls a core piece of its AI pipeline instead of renting it, and that usually means faster iteration and tighter integration with the rest of the product over time. For a small business in Jersey City or anywhere in the NYC metro already living inside Canva, the sensible response is the same one we give for every new AI design feature: use it for volume and speed, and keep a human-built, trademark-clearable brand identity as the filter everything else has to pass through. If you want that foundation built once — so it holds up whether the next asset comes from Craft, Midjourney, or a real camera — that's exactly the work our brand systems team does.