Canva Craft for AI video creators

Canva's first in-house image model (October 22, 2026) plans the picture before it paints it — 2K, 30+ aspect ratios, legible text in 55 languages. Here's why that matters for thumbnails, storyboards, and keyframes.

Last updated October 24, 2026 · Released October 22, 2026

What was announced

On October 22, 2026, Canva released Craft — the first image generation model built and trained from scratch by Canva Research, its in-house AI lab, with contributions from researchers of Leonardo.ai (acquired 2024) [official]. It is available to all users in Dream Lab and through Canva's public API (model craft_v1_2k), with MCP support for chatbots planned [official].

The architecture is the story: most image models turn a prompt straight into pixels, but Craft first writes a detailed plan of the composition — which elements appear, where they sit, their relative size, visual hierarchy, and any text — then renders it [official]. Canva trained it on its library of human-designed templates (posters, infographics, menus, typography-heavy layouts) rather than photographs [official].

Why video creators should care

Three jobs in every video workflow are stills, not video: the thumbnail that decides the click, the storyboard that decides the shoot, and the keyframe that decides what an image-to-video model animates. Craft's strengths land on exactly these:

Thumbnails and title cards. Text in images is where most generators embarrass themselves; Craft writes legible text in 55 languages and keeps it editable via Magic Layers, which converts the flat image into separate design elements [official].

Storyboards and keyframes. The planning stage means objects land where you asked and one element can be edited without redrawing the whole frame [official] — useful when a keyframe needs a specific composition for the video model to preserve.

Aspect ratios. 30+ ratios plus custom sizes [official] cover 16:9 openers, 9:16 verticals, and square posts from one generation.

The honest limits

Craft is a design model, not a photorealism model — it was trained on templates, so don't expect it to win at cinematic stills [official]. The DesignArena rankings Canva cites (4th overall, 1st in People & Portraits and Landscapes & Nature) are company-cited, not independently assessed [reported]. And the faster, cheaper Craft-fast / Craft-ultrafast variants are announced but not shipped [official] — if your pipeline is cost-sensitive, wait for those before committing.

FAQ

What is Canva Craft?

Canva Craft is Canva's first image generation model built and trained from scratch by Canva Research, released October 22, 2026 [official]. Unlike most image models that turn a prompt straight into pixels, Craft splits the job: a planning model first writes a detailed composition plan (elements, spatial relationships, hierarchy, text), then a rendering model paints it — up to 2K resolution in 30+ aspect ratios [official].

Why should AI video creators care about an image model?

Because every image-to-video workflow starts with a still. Thumbnails decide clicks; storyboards decide shoots; keyframes decide what image-to-video models animate. Craft's two advantages map directly onto those jobs: legible text in 55 languages (title cards that don't embarrass you) and layout planning (objects where you asked, targeted edits without redrawing everything) [official].

How does it compare to other image models for video work?

On DesignArena, Craft debuted fourth overall among image models, ranking first in People & Portraits and Landscapes & Nature [reported — company-cited, not independent]. Its differentiator isn't photorealism but design structure: it was trained on human-made templates (posters, infographics, menus), and Magic Layers can convert its output into editable design elements [official]. For thumbnails and title cards, that beats a prettier but uneditable image.

Is Canva Craft free?

It is available to all Canva users inside Dream Lab under Canva's existing plans [official]. API access (model craft_v1_2k) is billed through Canva's platform — check their pricing page for current rates. Craft-fast and Craft-ultrafast variants are planned but not released [official].

Can I use it with image-to-video models?

Yes — that's the workflow this guide is about. Generate a keyframe or establishing shot in Craft (16:9, 2K), then feed it to an image-to-video model as the opening frame. Craft's composition planning helps when you need a specific layout the video model should preserve.

Comparing the video models?

Stills are half the pipeline. See how the video models stack up — official specs, tester reports, and unknowns marked separately.

All comparisons