Canva launches Craft: ad images with text, made right inside Canva

On October 22 (US time), Canva launched Craft, the first image model built in-house by its Canva Research team, and opened it to every Canva user worldwide inside Dream Lab.

Craft is built for images where text and layout have to be tight

Craft works in two steps. First, one model reads the prompt and drafts a layout plan: which elements go in, where they sit, how big they are, and what the text says. Only then does a second model render the image, at up to 2K resolution and in more than 30 aspect ratios.

According to Canva, Craft can write text in 55 languages, and its images can be split into layers with Magic Layers, so you can fix one detail without regenerating the whole thing. The announcement doesn't list those 55 languages, so if your ads use accented text such as Vietnamese, test it yourself before anything goes live.

JJ Fiasson, Canva's head of generative AI, put it this way in Canva's announcement: "Craft delivers exceptional results for jobs where the layouts and text need to be perfect." For a store, that means sale banners, social posts with a price and a discount code, and product launch posters. These are also the first use cases Canva lists for Craft.

For product shots, put Craft's output next to a real photo

Canva points to the DesignArena leaderboard, a platform that compares AI-generated design images. Craft ranks 4th overall and tops the portrait and landscape categories. In the product category, though, it also sits at 4th, its weakest result of the four categories Canva published.

Read that way, models, settings and scenery are where Craft is strongest, while close-up product shots deserve a careful look. Our take: the safe split is to let Craft handle the background, layout and text, and keep real photography for the product itself. Shoppers buy off the image, so a render with the wrong color or a wrong detail can turn into complaints and returns.

Making more variations faster won't tell you which one sells

Ad people see AI in two ways right now. An X account that shares AI-made ads argues that the reason to use AI isn't that it's cheaper, it's that production stops being the bottleneck and your ideas become the only limit.

On the other side, Aniruddha Mishra, Director of Growth at direct-to-consumer health brand Hume Health, wrote on LinkedIn that anyone can ask ChatGPT or Claude for 20 ad concepts and then knock out a few static ads in Canva. In his view, the hard part is somewhere else:

"The barrier to entry for becoming a Creative Strategist is extremely low. The barrier to becoming a profitable, data-driven Creative Strategist is not."

Once a new image in Canva costs you one more prompt, the work that deserves investment shifts to picking which variations to run and reading the results. Rather than swapping every ad for new images, you could try this:

• Pick a static image ad you've already run and have results for
• Recreate it with Craft, keeping the message and the offer, changing only the visual
• Proofread the text in the image, especially diacritics, prices and product names
• Run both versions side by side on the same budget, then compare click-through rate and cost per order

If the Craft version doesn't win, you've spent very little and still learned which kind of image works for your customers.

What to keep an eye on

Canva says two more versions, Craft-fast and Craft-ultrafast, are on the way, aimed at high quality at a lower cost. Craft is also coming to the Canva MCP Server, so AI assistants can generate images as part of longer workflows. For agencies, Craft is already available through a public API, which means teams building marketing tools or ad generators can plug it straight into their own products.

Until then, one test on an old ad is enough to show which part of your store's ad images Craft should be making.