Canva Built Its Own AI Image Model. What Does Craft Change for Creators?

The image looks good. Now it needs a headline, a logo, the correct offer and a version that fits an Instagram Story without cutting someone's head off.

For a small business or solo marketer, that second half of the job can matter more than the first. Generating something impressive and getting a usable campaign asset out the door are different milestones.

Canva has introduced Craft, a dedicated image-generation model developed by Canva Research. Its pitch starts with the things a designer cares about: where objects sit, how text fits and whether the picture works in the space available.

Canva already has a tool that turns images from ChatGPT and Gemini into editable designs. Craft therefore has to earn its place by giving creators a useful starting image, with less work needed afterward.

The short answer: if you already make graphics in Canva, Craft is worth trying on your next one. Keep the image generator you trust while you see how much work Craft saves.

This early assessment draws on product documentation and launch reporting. We have not yet tested Craft ourselves.

What has Canva actually built?

In its October 22 announcement, Canva describes two stages: planning the composition, including placement and text, then rendering the image. It says training emphasizes design templates, with output up to 2K and about 30 aspect ratios.

For a promotional graphic, composition is the job. A striking background is less helpful if the main subject occupies the space where the offer needs to go. Building the model around design gives Canva a way to address that problem before the image reaches the editor.

Canva told TechCrunch it wanted greater control over quality and speed. Researchers from Leonardo.ai, the image-generation company Canva acquired in 2024, contributed.

This builds on an existing mix of technology. Canva's earlier AI product description already combined in-house models, Leonardo's Phoenix and partner models. Dream Lab was already part of that experience.

The names describe different pieces:

Canva's AI 2.0 announcement describes its Design Model as understanding design structure and producing layered output. Craft makes the image; Canva's other tools can take it further.

Start with the job you need to finish

Imagine a neighborhood bakery promoting a weekend pastry box. It needs a feed graphic, a Story and a slide for a local business presentation.

An illustrative brief might be: “A warm bakery scene with a pastry box in the lower half, morning light and clear space above for a short headline.”

The useful goal is a composition with room to work. You can then add the real business logo, approved wording and correct price as elements you can change. Next weekend's offer should not require persuading an image model to spell a different number.

Canva's Dream Lab instructions direct users to the homepage, then More and Dream Lab. Describe the image and select a result for a design. The page also covers reference inputs and follow-up edits, though it does not say which of those controls work with Craft.

For the bakery, a sensible sequence is to choose the visual direction, assemble the promotion, inspect it at phone size and adapt the layout for each destination. A wider crop may suit the slide; the Story may need the image moved lower to leave room for the offer.

That is where convenience could become meaningful. If the first result leaves fewer awkward corrections, there is less reason to start in another service. If you spend the afternoon repairing it, keeping everything in one account has not saved much.

Editable is the part worth understanding

Text drawn into an image is still part of the image. It does not automatically behave like a text box when a client asks to change three words.

Canva's Magic Layers guidance describes converting a flat design into a layout where text, colors and elements can be adjusted. Treat generation and conversion as separate steps, then inspect the result before handing it over.

Magic Layers also works with images from other generators, including ChatGPT and Gemini. An external model can supply the visual while Canva handles the design work afterward. That makes the quality and convenience of Craft's first image central to the choice.

There is one export detail to check before promising a finished campaign asset. Dream Lab's help page says generated images carry a visible AI label, but does not explain whether that label remains in downloaded files. Check the actual export before client delivery. Canva's AI terms also prohibit removing provenance metadata or presenting AI output as human-made.

When another image tool still makes sense

Keep the tool that solves the difficult part of your job. If a generator reliably handles a particular style, product or character, that result deserves weight when deciding whether to switch.

These are reasons to consider a workflow. The best fit depends on how well it handles your own material and how much correction the finished design needs.

Access and allowance: check the actual account

Canva announced Craft for all users through Dream Lab. Organization administrators can restrict access. Availability also tells you little about the number of images your account can make, and the announcement does not establish Craft as Dream Lab's default model.

Canva's AI usage guide gives best-case monthly examples of up to 20 Premium uses on Free, 200 on Pro and 400 on Business. Those are shared-plan estimates, not guaranteed Craft image counts. Craft's exact tier and debit are not established by that page. Generating images, retries and Magic Layers can draw on your allowance. Limits reset monthly; behavior after exhaustion depends on the plan and tool tier. Check Settings → Billing → AI usage, and leave room for the work after the first prompt.

For commercial work, Canva's AI terms say that, between you and Canva and to the extent the law permits, you own your output. Output incorporating or modifying Canva library content is subject to its license. They do not guarantee unique output or clear third-party rights. Check your source material, logos and final claims, and review privacy settings before uploading confidential client work.

What is still coming?

Canva plans faster Craft variants and Craft-specific access through its MCP connection to AI assistants. Existing Canva integrations do not prove those Craft features are available today.

Developers will find craft_v1_2k in Canva's API model overview. The image API remains a preview: public apps built on it cannot pass Canva's review, and each job uses the user's AI allowance.

The Choosely verdict

Craft's most interesting test will happen after someone likes the first image.

Can they finish the post, correct the offer, make the second format and hand over something another person can edit? That is the standard that matters to a busy creator.

For an existing Canva user, a small project is a sensible place to assess Craft once access and usage are clear. Keep the wording and brand assets under your control, inspect the export and compare the work needed to reach an acceptable result.

The reason to pay attention is the possibility of a shorter trip to a finished design. Craft earns a permanent place in that workflow when it makes the whole job easier.