Canva released Craft, its first internally built AI model for image generation. The model is now available to every user on the platform without extra tiers or fees, and the company states that it delivers stronger customization for design work at lower cost than the external models Canva previously used.
Prior setup and the shift
Before this release Canva depended on third-party image models to power its AI features. Those external systems were trained primarily on photographic data rather than the graphic elements, layouts, and brand assets that dominate most Canva projects. Craft changes that dependency by bringing the model in-house and training it specifically for design tasks.
The October 22, 2026 release marks the first time Canva has shipped an image model developed entirely by its own teams. TechCrunch reports that the model is being rolled out to the full user base immediately. Thurrott notes the same date and emphasizes that the training focus was design output rather than photorealism.
Technical positioning and availability
Craft is presented as a tool that fits inside Canva’s existing editor rather than requiring users to export work to another service. The company claims the model supports more direct adjustments to generated images while they remain inside the canvas. No public details on model size, training data volume, or benchmark scores have been released.
Because the model is offered to all accounts at no added charge, teams that already pay for Canva subscriptions can test it without signing separate contracts or monitoring API usage. The absence of tiered access removes one common friction point that appears when companies introduce new generative features.
Reactions and open questions
No independent third-party benchmarks or user studies have appeared yet. The two available reports rest on Canva’s own statements about performance and cost. Where the sources differ in emphasis, TechCrunch highlights broad availability while Thurrott stresses the design-specific training approach. Both accounts agree on the release date and the fact that this is Canva’s first in-house image model.
Why it matters
Design teams that live inside Canva now have an integrated generation option that does not force them to leave the editor or pay separate bills for each image. The stated focus on design elements instead of photographs could reduce the number of follow-up edits needed for presentations, social posts, and marketing collateral. That efficiency gain, if real, compounds across thousands of daily users who currently generate an asset, then spend additional minutes aligning colors, fonts, or layout.
The cost claim rests entirely on Canva’s assertion rather than published pricing comparisons. Users will still need to run side-by-side tests against the third-party options they used before to judge whether output quality justifies the switch. If the model meets the stated standard, Canva gains tighter control over its feature roadmap and reduces exposure to changes in external providers’ terms or pricing.
Longer term, an in-house model gives Canva the option to iterate on capabilities that matter most to its core audience without waiting for outside vendors to prioritize the same use cases. Teams that standardize on Canva for both design and generation may find fewer handoff points and fewer accounts to manage. The practical test will be whether daily output improves enough to offset any remaining limitations in photorealistic or highly complex scenes.
The move also signals that Canva intends to treat generative image tools as core infrastructure rather than a bolted-on service. That stance affects how quickly new controls or brand-guard features can reach users and how much visibility the company has into usage patterns inside its own platform.
---
Sources:
{}
No comments yet