Canva launches Craft, its first in-house image model, with two-stage design planning and 2K output

Canva has released Canva Craft, its first in-house AI image generation model, to all users globally. The model uses a two-stage architecture that plans a design before generating the image. Canva says this gives "greater precision, consistency and control." Canva announced the model on Thursday, October 22, 2026.

Key specifications

• Max resolution: 2K
• Aspect ratios: more than 30
• Architecture: two-stage, with a planning step followed by image generation
• Training data: Canva says it trained Craft on "millions of human-made templates" from its content library, including social media posts, presentations, infographics, and menus
• Availability: all users globally, through Dream Lab (three-dot option in Canva's left-hand menu)
• Pricing: not yet disclosed for any separate or API access
• Parameter count, context window, training cutoff: not disclosed

How the two-stage design works

According to Canva, planning the design first helps the model handle images that need legible text and precise object placement. Both are common failure points for image generators. Canva has not published technical details of the planning stage or how it connects to the generation stage.

Beyond social and marketing content, Canva says Craft can produce realistic scenes, digital art, and environmental imagery.

Benchmark position

Craft currently ranks fourth on Design Arena's leaderboard, a crowdsourced benchmark for AI-generated designs, according to The Verge's reporting. It sits behind three OpenAI models:

1. GPT-Image-2.5 Sunburst
2. GPT-Image-2.5 Flare
3. GPT-Image-2

No numerical scores were provided. The ranking is a snapshot of a crowdsourced leaderboard, not an independent evaluation, and it can shift as votes accumulate.

Safety and roadmap

Canva says Craft includes "robust input and output moderation, privacy and safety controls, and reporting tools." It has not detailed how these work.

The company plans to:

• Let AI agents connect to Craft through the open Model Context Protocol (MCP)
• Release more efficient Craft-fast and Craft-ultrafast variants. No dates have been given.

Context

Craft follows Canva's earlier AI push. In April, the company launched prompt-based editing, which lets users create or change a design by describing it. It also added layers to AI-generated images, so users can edit part of an image without altering the rest.

What this means

Canva is moving from integrating third-party image models to shipping its own. That gives it control over cost, latency, and behavior on its core use cases, which are layouts, text-heavy graphics, and marketing assets. Training on Canva's template library fits the product's strengths, though the company has not shared how the data was licensed or curated.

The fourth-place Design Arena ranking is competitive for a first release, but OpenAI holds the top three spots. Craft's differentiation depends on the planning stage actually delivering better text rendering and object placement in real workflows. Canva has not published evidence for that beyond its own claims.

The planned MCP support and the faster Craft variants are worth watching. If Canva exposes Craft to agents and releases cheaper tiers, it could compete as a design-specific generation backend rather than only a feature inside its own app. Pricing and API availability are the open questions.