Canva Launches Craft, Its First In-House Image Generation Model

Canva has launched its first in-house image generation model, Canva Craft, trained on the design platform's own library of human-designed templates (posters, social posts, infographics, menus, typography-rich layouts) rather than the photograph-heavy datasets behind most rival models. TechCrunch reported the model runs in two stages: a planning pass that fixes spatial relationships, scale, visual hierarchy and text, then a render pass that outputs up to 2K across 30 aspect ratios.

It is live to all Canva users in the Dream Lab tab, exposed via the Canva API, and reachable from ChatGPT and Gemini through Canva's MCP connector. Cheaper variants, Craftfast and Craft-ultrafast, are planned.

Preliminary results put Craft fourth on Design Arena's independent image-generation leaderboard, ahead of entries from Google, Microsoft, Meta, xAI and ByteDance, which Capital Brief says makes Canva "the second-highest-ranked AI laboratory for image generation behind OpenAI." The launch is the first public release from Canva Research, a team Capital Brief reports has tripled in size over the past year, and follows the 2024 acquisition of Leonardo.ai, whose researchers contributed to the work.

Capital Brief frames the move as a cost play. Higher-than-expected AI usage costs earlier in 2026 forced Canva to cut its revenue growth target from 30% to 20%, and owning the model stack trims third-party API spend while opening a licensing channel for outside developers. The release drops into a crowded stretch of image-model news, with our generative-AI tracker logging 314 stories in the last 90 days, though Craft is only the eighth entry on our narrower AI-art beat across that window.

Neither TechCrunch nor Capital Brief publishes API pricing, per-task benchmark subscores, or direct quotes from named Canva executives.