Canva Craft: how to keep editing after image generation

A balloon, a background and a heading can share one image. Continuing the design depends on selecting an element separately. Canva’s Magic Layers illustration places a purple selection box around the balloon: the intended control is over an object, color or text, rather than only the complete flat image.

Changing a poster’s date is a useful example: retaining a picture you like while changing its lettering is different from generating another similar poster. Magic Layers documents selection, movement, deletion, independent resizing and replacement. It also accepts images from other generators. Continued editing therefore depends on the design tool that takes over, as well as the model that made the picture.

• See the official element, color and text illustrations ↗Official feature illustrations, not an independent matched-input test.

Introduced on October 22, Craft is described by Canva as Canva Research’s first image foundation model trained from scratch. Canva separates composition planning—elements, relative scale, hierarchy and lettering—from rendering an image at up to 2K. This published architecture explains how the company organizes generation; the usefulness of its composition and text still depends on actual outputs.

The public generation job returns an image download and, optionally, an uploaded image asset. Placing that asset into a design can still place it as one flat image. Converting its contents into separately editable elements is another task. The reference distinguishes these uses: ordinary Create design placement uses no AI allowance, while Magic Layers image-to-design import starts a new AI conversion.

• Read Craft’s original architecture and access account ↗

The result types make the distinction concrete: generation supplies an image; a successful image-to-design job supplies a Canva design and editing link. Conversion uses an image asset already in the account and checks design-write permission and image-to-design capability. Generation has its own asset-write permission and capability. Access to one does not establish access to both. A developer connecting the stages must handle both capabilities and each job’s result.

Constructing an editable design from a raster image does not guarantee restoration of the author’s fonts, hidden elements or original project structure. The help page favors distinct graphic elements, illustrations and stylized visuals. A single-page PNG or JPEG is a common starting point; the UI also lists a 5000px longest-edge limit. Photographic separation may vary. The API lists additional raster formats, so the UI’s shorter list is not a universal input specification.

• Compare image-to-design inputs, permissions and allowance ↗
• Check UI input formats and editing scope ↗

Access needs the same separation. Canva announces Craft in Dream Lab, while the generation and image-to-design APIs remain previews that prevent public apps or integrations using them from passing general distribution review. Starting generation and conversion each uses account AI allowance; polling the conversion does not. Eligibility, balance and debit amounts cannot be inferred from an illustration. Craft MCP remains coming soon in the announcement.

With a small poster to revise, inspect the change you actually need: correct characters and spacing after replacing a date, plausible background and overlap after moving an object, and preservation of the elements you already like. The publisher’s 55-language text claim does not replace checking a complete Chinese poster. The value of continued editing lies in these small changes that can be selected and inspected.

• Continue from the official introduction to Craft access ↗
• Check the shared AI allowance conditions ↗