Canva released Canva Craft, the first image-generation foundation model trained from scratch by its own research team — tuned for legible text and layout-aware design output rather than photorealism.
2026-10-22 · by Moe Ameen
Canva released Canva Craft on October 22, 2026 — the first image-generation foundation model the company says was built and trained from scratch by Canva Research. Unlike general image models tuned for photorealism, Craft is aimed at design output: posters, social posts, presentations, infographics, and typography-heavy layouts with legible, correctly placed text. Co-founder and COO Cliff Obrecht called it "a huge moment" for the company, which is positioning Craft against image generators from Google, OpenAI, ByteDance, and Meta.
The model works in two stages. It first interprets the prompt and drafts a plan — mapping spatial relationships, relative scale, visual hierarchy, and the text that will appear in the image — before generating any pixels, an approach Canva frames as "less about pixels and more about planning." It outputs images at up to 2K resolution across a wide range of aspect ratios, with editable, controllable results. Canva says the model was trained on millions of human-made templates from its own content library, which it credits for the stronger text rendering and layout awareness versus photo-trained models.
Craft is available now via the Dream Lab tab in the Canva app — described at launch as open to all Canva users globally, with Canva's usual free/paid credit tiers likely applying to heavier use — and to developers through Canva's public API; a Canva MCP Server that lets AI agents call the model is described as coming soon. Canva also says it will soon release two cheaper variants — Craft-fast and Craft-ultrafast — pitched as high-quality generation "at a fraction of the cost of leading frontier models." On the independent DesignArena leaderboard, Canva reports Craft debuted fourth, behind three OpenAI models (GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare, and GPT Image 2) and ahead of entries from Google, Microsoft, Meta, xAI, and ByteDance — making Canva, by its own account, the second-highest-ranked lab for image generation behind OpenAI. Treat the ranking and cost claims as early and company-reported; verify current specs and pricing on Canva's own pages.
The practical move the day Craft ships is to treat it as a better source of still assets and wire it into a pipeline that actually publishes. Craft's strength is the hard part of marketing imagery — readable text and intentional layout — so it's an unusually good way to generate hero images, backplates, and poster-style graphics. [Kompozy](/) is the layer that turns those stills into a published week: drop a Craft image into Kompozy and it becomes brand-exact Carousel Posts, Photo Posts, and Quote Graphics rendered through [HyperFrames](/glossary/hyperframes), all held to one voice and banned-word list by your [Persona Brief](/glossary/persona-brief), then scheduled and fanned across the eight social platforms plus blog and email on [Autopilot](/glossary/autopilot) behind a per-post review gate.
That also covers what a still model can't. Because Kompozy is a full generation-and-publishing engine, the same campaign that starts with a Craft graphic can carry into face-locked [Persona Shorts](/glossary/persona-shorts), vertical [Clipped Shorts](/glossary/clipped-short) pulled from a long video, a Blog Article, and an Email Newsletter — one idea, every format, in a single voice. Craft makes the picture read; Kompozy makes it a scheduled, on-brand presence everywhere your audience is.
Canva Craft is Canva's first proprietary AI image-generation model, announced October 22, 2026, and built from scratch by Canva Research. It is tuned for design output — posters, social posts, presentations, infographics, and typography-rich layouts with legible text — rather than photorealism, and generates images at up to 2K resolution across a wide range of aspect ratios.
Craft is reachable via the Dream Lab tab in the Canva app — reported at launch as open to all Canva users globally, with Canva's usual free/paid credit tiers likely applying to heavier use — and to developers through Canva's public API. Canva has said it will soon release cheaper Craft-fast and Craft-ultrafast variants "at a fraction of the cost of leading frontier models," but has not published final pricing — confirm current terms on canva.com.
Most image models are trained mainly on photographs; Canva says Craft was trained on millions of human-made design templates, which it credits for stronger in-image text and layout awareness. It also plans the image (spatial relationships, scale, hierarchy, and text) before generating, rather than going straight to pixels.
No. Craft is an image-generation model — it makes stills. It does not generate or clip video, write captions in a governed brand voice across formats, assemble finished multi-format posts, or schedule and publish to social platforms. A content engine like Kompozy takes a Craft image and turns it into on-brand posts scheduled across the eight social platforms plus blog and email.