Canva launches Canva Craft image generation tool

Canva Craft Launches Image Generation Tool

Canva launched Canva Craft on Thursday, marking the design platform’s entry into the competitive AI image generation space alongside models from Google, Meta, and OpenAI, Business Insider reported. Users can access Craft through Canva’s Dream Lab, including free accounts subject to standard usage limits, provided they sign into an active Canva account.

The Tech TL;DR:

• Canva released its first image-generation model, Canva Craft, accessible through Dream Lab for both free and paid accounts on Thursday.
• The model plans layouts, proportions, and text before generating images, utilizing human-designed templates rather than photographs for training.
• In performance testing against ChatGPT, Muse, and Gemini across three prompts, Canva delivered mixed results, struggling with basic illustrations while handling composition on complex scenes.

Canva Craft Uses Templates to Plan Layouts

The underlying architecture of Canva Craft approaches image generation by planning a design’s layout, proportions, and text prior to pixel rendering. This design-first pipeline aims to produce legible lettering and accurate compositional details. According to the release announcement, the model was trained on human-designed Canva templates—encompassing posters, menus, and presentations—rather than raw photographic datasets.

Evaluating Performance Across Three Difficulty Tiers

Business Insider tested Canva Craft against competing models using three prompts with escalating complexity. The initial test involved a simple birthday card illustration for a cartoon character named Jessica with specific physical attributes. Canva fell short on this prompt, producing an awkward font and an oversized head while omitting requested details like paisley.

By comparison, ChatGPT and Muse interpreted the card prompt successfully, while Gemini produced a literal image of an actual card. For the second task involving object extraction—placing a red garnet pendant from a hand into a showroom vitrine—Canva correctly identified the pendant but positioned it floating in the air, defying physical laws much like Gemini’s output. ChatGPT and Muse demonstrated better adherence to real-world physics for that task.

The third test required generating a photorealistic long-haired orange tabby cat in a Halloween costume with bat wings perched on a gothic medieval castle. Canva’s output featured strandy and high-contrast fur, but the composition contained no missing or malformed components. ChatGPT emerged as the clear winner for that tier due to realistic lighting and white balance.

Canva Craft Has Room for Growth

Evaluations indicate that Canva Craft retains significant room for growth compared to established standalone models, particularly given its heavy focus on graphic design workflows.