Log in Sign up
Business Formic Raises $50 Million to Move Robots-as-a-Service From Single Tasks to General-Purpose Machines Security Datadog Offers Nearly $4 Billion for Upwind, a 5 Percent Premium Over the Startup's Own Private Round Productivity Canva Trained Its First Image Model on Design Templates Instead of Photos, and Says the Text Finally Reads News Rivian Buys a 15-Person UK World-Model Startup to Simulate the Edge Cases Its Fleet Has Never Filmed Business Ex-OpenAI Science Chief Kevin Weil Raises 215 Million for Robot Chemistry Labs at an 865 Million Valuation AI Agents Google Gives Every AI Agent Its Own microVM: 200,000 Agents, 500 Wake-Ups a Second ChatGPT Uber Puts All 30,000 Employees on ChatGPT Work After a Pilot Where 80 Percent Were Not Engineers News Microsoft Says Its Diagnostic Orchestrator Beats Doctors 83 to 36 Percent on Hard Cases, With a Catch Business Formic Raises $50 Million to Move Robots-as-a-Service From Single Tasks to General-Purpose Machines Security Datadog Offers Nearly $4 Billion for Upwind, a 5 Percent Premium Over the Startup's Own Private Round Productivity Canva Trained Its First Image Model on Design Templates Instead of Photos, and Says the Text Finally Reads News Rivian Buys a 15-Person UK World-Model Startup to Simulate the Edge Cases Its Fleet Has Never Filmed Business Ex-OpenAI Science Chief Kevin Weil Raises 215 Million for Robot Chemistry Labs at an 865 Million Valuation AI Agents Google Gives Every AI Agent Its Own microVM: 200,000 Agents, 500 Wake-Ups a Second ChatGPT Uber Puts All 30,000 Employees on ChatGPT Work After a Pilot Where 80 Percent Were Not Engineers News Microsoft Says Its Diagnostic Orchestrator Beats Doctors 83 to 36 Percent on Hard Cases, With a Catch

Canva Trained Its First Image Model on Design Templates Instead of Photos, and Says the Text Finally Reads

Productivity · By: AIReport Admin · 2 hours ago · 👁 11
Canva Trained Its First Image Model on Design Templates Instead of Photos, and Says the Text Finally Reads

Canva released Craft, its first in-house image generation model, trained on human-designed templates rather than photographs. It plans a layout first, then renders up to 2K resolution in 30 aspect ratios, and is available to all users through Dream Lab and the Canva API, with ChatGPT and Gemini access planned via MCP. Cheaper Craft-fast and Craft-ultrafast variants are coming soon.

Canva has released Craft, the first image generation model developed by its in-house research group, Canva Research. The model is available to all users starting Thursday, reachable from the Dream Lab tab, through the Canva API, and the company says it plans to expose it inside third-party chatbots such as ChatGPT and Gemini through Canva's MCP integration.

Craft works in two stages. The first stage reads the prompt and produces a plan: how elements sit in space, their relative scale, the visual hierarchy, and where text goes. The second stage renders an image of up to 2K resolution in one of 30 aspect ratios. Canva says that split is what keeps object detail accurate and text legible, and that users can edit parts of an image without disturbing the rest.

The most distinctive choice is the training data. Instead of photographs, Craft was trained on human-designed templates from Canva's own library: posters, social posts, presentations, infographics, menus and typography-heavy layouts. That means a generated image can be converted back into a layered design with Canva Magic, the image-to-design transformer the company shipped last year, which is a workflow pure photo-trained models cannot offer.

There is a business logic underneath. Canva has been routing generation work to third-party model providers, and owning the model cuts those API bills while giving the platform control over quality and speed. Researchers from Leonardo.ai, the media generation company Canva acquired in 2024, contributed to Craft. Two cheaper variants, Craft-fast and Craft-ultrafast, are coming soon.

The model lands in a crowded field: Google, OpenAI, ByteDance and Meta all ship image generators, and Canva told the Australian Financial Review it ranks above Google's model and behind only OpenAI's. That ranking is a company claim, not an independent benchmark, and no third-party evaluation has been published yet.

The timing is not accidental. Canva cut its growth forecast for 2026 after shelving heavily promoted AI products in August because usage costs threatened to overwhelm the business, and reports on the resulting valuation markdown disagree: TechCrunch put the fall at $42 billion to $35 billion, while the Australian Financial Review described backers cutting $10 billion and the valuation landing near $50 billion. The company was tracking toward $4 billion in revenue last year.

The bet here is that design intent beats photorealism for the audience that actually pays: marketers, small businesses and teams who need a flyer where the discount code is spelled correctly. If Craft's template-native training holds up in production, Canva gets margin control and a moat made of layout, not pixels.

Comments (0)

Log in to join the discussion

Log In

No comments yet