Canva launches in-house image model Craft, opening access through Dream Lab

Canva has introduced Craft, its first image generation model trained in-house from scratch, and says it became available to all Canva accounts globally through Dream Lab starting Oct. 22. The company says Craft uses a two-stage process: a planning model first interprets prompts and maps out composition, object placement, relative size, visual hierarchy, and text, then a drawing model renders the final image at up to 2K resolution across more than 30 aspect ratios. Canva also says generated images can be broken into editable layers with Magic Layers. The launch comes as Canva faces pressure to reduce AI serving costs. Startup Daily, citing an August shareholder letter, reported that CEO Melanie Perkins said the company had become too dependent on frontier models from providers such as OpenAI and Google, pushing average AI task costs too high. Canva also lowered its annual revenue growth forecast from 30% to 20%. The company said Craft was trained on millions of human-made Canva templates rather than mainly on photos, and noted that future integrations through its MCP server could let assistants such as ChatGPT and Gemini call the model. Pricing, free usage limits, and the rollout dates for cheaper versions called Craft-fast and Craft-ultrafast have not been disclosed.