Canva launches Craft, its own image generation model
Canva’s new model plans compositions before rendering images and is available through Dream Lab and an API.
Canva says Craft plans an image’s composition before rendering it: the plan covers objects, their relative size, placement and text. The company built the model for design work, including social posts and marketing material. Users can access it through Dream Lab; the announced Craft integration with Canva MCP is still described as coming soon.
For developers, the request documentation lists Craft as craft_v1_2k. Starting a job needs a text prompt, an idempotency key identifying the request, and a user access token with asset:write permission; the user also needs the image_generation capability. Generation consumes that user’s AI credits. The API remains a preview: breaking changes may arrive without a new version, and public apps using it cannot pass review.
The result endpoint lets only the creating app and user retrieve the job. It may need repeated queries until success or failure. A successful result supplies a download link valid for 24 hours. Saving to the user’s uploads is optional; when requested, the job remains in progress until that import finishes, so the returned asset is ready to use.
The model guide explains selection: pinning a model prevents substitution, and the job fails if it cannot serve the request. Omitting the selection lets Canva choose. Completed jobs and their image download links remain available for 24 hours.
The request key is retained for 24 hours from job creation. During this window, reusing it returns the original job without duplication, even with changed parameters.
The Decoder also reported the launch, summarizing Canva’s announcement.
