THE ESSENTIALS IN 30 SECONDS

  • Craft uses a two-stage design: a planning step maps layout, scale, hierarchy and text, then a renderer draws the image.
  • It outputs up to 2K resolution in more than 30 aspect ratios and is available to all Canva users via Dream Lab and through Canva's API.
  • Cheaper Craft-fast and Craft-ultrafast variants and MCP support are announced but undated.

Canva, the Australian design platform, has released Craft, the first image generation model developed by its own research team. Announced on October 22, the model is pitched less at photorealism than at the kind of images Canva users actually make: posters, social posts, menus and other layouts where text and composition matter.

A model that plans before it draws

Craft works in two stages. A first model interprets the prompt and produces a plan covering where elements go, their relative scale, the visual hierarchy and the text to include. A second stage then renders the image. Canva says this separation helps it produce legible text and make targeted edits without regenerating the whole picture.

According to TechCrunch, the model was trained mainly on human-designed templates from Canva's library rather than on photographs, and researchers from Leonardo.ai, which Canva acquired in 2024, contributed. TestingCatalog reports that Canva then applied supervised fine-tuning and reinforcement learning guided by its professional designers. Craft generates images up to 2K resolution, in more than 30 preset aspect ratios, and Canva says generated text can appear in 55 languages.

Availability

Craft is available to all Canva users through the Dream Lab tool and to developers via Canva's public API. Output can be converted into layered, editable designs with Canva's Magic Layers feature. JJ Fiasson, Canva's head of generative AI, told SmartCompany the model performs well on jobs where layout and text must be precise.

The company has also announced two cheaper variants, Craft-fast and Craft-ultrafast, plus support for its MCP server so that assistants such as ChatGPT or Gemini can generate images through Canva. None of these has a release date.

Context and caveats

Canva says Craft placed fourth on the DesignArena benchmark; that ranking is the company's own framing and leaderboard positions shift quickly. Craft enters a crowded field dominated by Google, OpenAI, ByteDance and Meta, and Canva has itself relied on third-party models until now. Building in-house is expected to lower its API costs, TechCrunch notes, at a time when the company has reportedly trimmed its 2026 growth forecast. Pricing for heavy API use has not been detailed in the sources reviewed.