AIINEWS

AIINEWS

Global intelligence on artificial intelligence and computing

Canva Craft: Canva's first photo creation model that plans your layout before drawing

AIINEWS Editorial Board•ai-machine-learning
23 October, 2026•2 min read
Canva Craft: Canva's first photo creation model that plans your layout before drawing

Executive Briefing

Synthesized and verified with gemini-2.0-flash-thinking-exp

“Canva just launched Canva Craft, the first image creation model built and trained from the ground up”

Key Takeaways

  • 1.Canva just launched Canva Craft, the first image creation model built and trained from the ground up by Canva Research's internal AI research lab.
  • 2.The model is available to all Canva users and to developers via API, competing directly with image generation models from Google, OpenAI, ByteDance and Meta.
  • 3.*Some sample photos created by Canva Craft.

Canva just launched Canva Craft, the first image creation model built and trained from the ground up by Canva Research's internal AI research lab. The model is available to all Canva users and to developers via API, competing directly with image generation models from Google, OpenAI, ByteDance and Meta.

Some sample photos created by Canva Craft. Photo: Canva

Two-phase architecture

Unlike models that generate pixels directly from the prompt, Craft separates composition from rendering:

  • Phase 1: a model plans to read the prompt and write a detailed description of the layout, including appearance elements, spatial relationships, relative proportions, visual hierarchy and text to be displayed.
  • Phase 2: a rendering model converts the plan into a maximum resolution image of 2K, with more than 30 aspect ratios.

Canva says this approach makes the model display text easier to read, places objects more accurately, and allows for partial editing without changing the rest of the image. According to the developer page, Craft receives prompts in all languages ​​and displays text in images in 55 languages.

Training with design instead of photos

According to TechCrunch, Canva uses human-made designs in its library for training, including posters, social media posts, presentation slides, infographics, menus and text-rich layouts. This makes images created with Craft suitable for turning into layered designs that can be edited using the Magic Layers feature. Researchers from Leonardo.ai, which Canva acquired in 2024, also participated in developing the model.

Ratings and usage

  • Canva said Craft debuted at No. 4 on DesignArena's photo creation rankings, topping the People & Portraits and Landscapes & Nature categories, and No. 2 in Abstract & Patterns.
  • Users access Craft in the Dream Lab section. Developers can call via the API with model name craft_v1_2k.
  • Canva plans to support image generation in third-party chatbots such as ChatGPT and Gemini via MCP.
  • Two cost-effective versions, Craft-fast and Craft-ultrafast, will be available soon.

Canva previously used many third-party image creation models, so its own model can help it reduce API costs and better control quality and speed. The move comes as Canva lowered its 2026 growth forecast, and according to outside analysts, the company's valuation is said to have dropped from $42 billion to $35 billion.

Source: TechCrunch, Canva Newsroom, Canva Developers, The Verge

Content and editorial synthesis copyright by AIINEWS Editorial Board
Share this intelligence report:

Related Dispatches