Canva has publicly released Canva Craft, its in-house image generation model, to compete with major tech rivals and is rolling out faster variants like Craft-fast and Craft-ultrafast to accelerate production and reduce costs.

Craft is accessible through Dream Lab and the public API, and its performance has earned it strong placements in DesignArena rankings, with ongoing plans for four image variations per prompt and editable results via Canva7s editor and Magic Layers.

In initial tests, Craft produced legible typography and better spacing on banners, though it could still misplace elements or introduce unintended text badges, highlighting ongoing areas for refinement.

Text handling is robust across languages, rendering readable on-image text in 55 languages with safeguards and editable elements via Magic Layers.

Craft supports four variations per prompt and enables editable results, though some workflow quirks were noted, such as unsolicited badges and an edit box acting more like an instruction field.

In a birthday-card comparison, Muse outperformed Craft by incorporating context and style, while Canva Craft fared poorly on details like an incorrect paisley motif; Gemini and ChatGPT also showed strengths.

Craft uses a two-stage plan-then-render process: a design planning stage determines object placement and text positioning, followed by image generation that supports up to 2K resolution and 30 aspect ratios.

Compared with ChatGPTs image tool, Craft delivered four images per prompt and generally matched or approached text accuracy, offering a calmer, designer-friendly aesthetic but still requiring proofreading for text and banner placements.

Overall, Canva Craft shows room to grow in image generation, with designers still preferring manual adjustments for fine-tuning in many cases.

Leonardo.ai researchers, acquired by Canva in 2024, contributed to Crafts development, signaling internal collaboration and investment in AI capabilities.

Craft is trained on millions of Canva templates—not photographs—across posters, menus, presentations, infographics, and other designs, enabling outputs that resemble human-made web banners with emphasis on text legibility and layout control.

Users can sign in to Canva, open Dream Lab, select Craft, input a prompt, choose resolution and aspect ratio, generate, and refine with Magic Layers, with admins able to disable the tool in settings.