Canva Craft plans the picture first. The useful test is what happens when you edit it

Canva's first image model, Craft, starts by planning a composition before it renders the picture. That design choice is more interesting than a launch-day leaderboard position: the company wants generated images to fit the work of arranging, revising and reusing a design, rather than end the process as a single attractive bitmap.

The October 22 announcement says Craft is available globally in Canva's Dream Lab and through a public API. It describes output up to 2K resolution and a separate planning stage that determines elements, their placement, scale and text. These are the company's architectural and product claims, not results from a GeekRadar benchmark.

Plan the relationship between elements

A poster can fail even when every individual object looks convincing. The title might compete with the subject, an important date can disappear into a busy background, or a crop can remove the visual connection between two elements. Those are composition problems, not simply failures to draw a hand or spell a word.

Craft's stated approach gives that arrangement an explicit stage. The subsequent renderer turns the plan into pixels. Separating the two could make targeted changes more predictable: move one element or alter its prominence without treating every revision as a completely new picture.

That is the useful hypothesis. The announcement does not establish that every requested edit will preserve every unaffected detail. A design still needs to be checked at its intended size, with the final wording, rather than judged only as a large preview.

The second half of the job is revision

Canva connects Craft to Magic Layers, its system for turning image content into editable elements. Its research-lab introduction explains why that matters to the company's broader strategy: iteration and format changes are part of design, not an afterthought once an image has been generated.

Consider a campaign that needs a square post, a narrow banner and a portrait flyer. A literal crop may lose the date or leave the subject squeezed against an edge. Repositioning separate components can preserve hierarchy across those formats. Whether a particular output supports that cleanly remains a property of the actual result, not of the word “editable” on a product page.

The same applies to text. Canva says Craft can create text in 55 languages. That expands the intended audience, but it does not certify spelling, grammar, cultural fit or the correctness of a price. A visibly polished error is still an error, and can become harder to notice because the layout looks finished.

Available now is different from coming next

The announcement names faster Craft variants and further editing capabilities as future developments. Its MCP section also labels that connection as coming soon, despite broader language earlier in the post about developer access. For now, the clear entry points described as available are Canva itself and the public API.

That distinction helps keep the launch useful. A designer evaluating the feature can concentrate on a real recurring task: creating a draft that survives the next request for a new size, revised date or different emphasis. There is no need to treat the promised future workflow as already delivered.

Canva's strongest opportunity is reducing the distance between “that looks good” and “I can actually use this.” The decisive revision may be the least glamorous one: changing a line of text without having to rebuild the image around it.