Canva launches Brush, its first vision-language model dedicated to design

On October 29, 2026, Canva, the Australian online design platform, unveiled Brush, its first artificial intelligence model combining vision and language, dedicated to graphic editing. Although there is no direct link to crypto-assets, this announcement illustrates the rise of proprietary AI tools in creative sectors, a movement likely to redefine competitive dynamics between centralized and decentralized solutions. Context Founded in 2013, Canva has established itself as a major player in digital design, with over 170 million monthly users in 2026, according to the company's internal data. Long dependent on external AI models, the company has accelerated its investments in artificial intelligence research, notably through Canva Research, its internal laboratory with 160 employees. The launch of Brush is part of this verticalization strategy, following that of Canva Craft, an image generation model unveiled a few weeks earlier. In the market for AI-assisted design tools, competition is intensifying. Players such as Adobe (with Firefly) or start-ups specializing in generative AI, such as MidJourney or Stability AI, dominate the segment. At the same time, blockchain projects like Ocean Protocol or Fetch.ai are exploring decentralized models for the monetization of creations or the management of copyrights. Canva's announcement comes in a context where crypto-assets, although stable (Bitcoin and Ethereum showing variations of less than 3% over seven days, according to CryptoScore Pro), remain sensitive to technological innovations likely to modify usage. The Facts Canva officially presented Brush on October 29, 2026, via a blog post and an article published on CryptoBriefing. Here are the key elements of the announcement: Features : Brush is a vision-language model (VLM) designed to edit designs based on textual instructions. It interprets layouts, layers, and visual styles, allowing modifications ranging from minor touch-ups to a complete redesign of a project. According to Canva, the generated outputs remain fully editable by the user. Performance : The company claims that Brush reduces editing costs by 46% and accelerates tasks by 69% compared to its previous AI agents. The announced latency is less than 2 seconds per request, according to internal tests. However, no external audit has been conducted to confirm these figures. Monetization : Canva plans to offer Brush via a paid API, charged at 0.005 USD per generated image. The model will first be accessible to Pro users before a wider rollout. No integration with blockchain protocols or crypto-assets is mentioned. Development : The model was trained on a proprietary corpus of millions of images and descriptions, including anonymized user data. Canva emphasizes that Brush relies on a multimodal transformer architecture, combining an image encoder and a text decoder. No details were provided regarding the number of images generated during tests, nor the precise criteria for measuring time and cost savings. Furthermore, the CryptoBriefing article indicates that Canva is working on Brush's interoperability with external AI platforms, without specifying which ones. Understanding Key Concepts Vision-Language Model (VLM) A vision-language model (VLM) is an artificial intelligence system capable of simultaneously processing visual data (images, graphics) and textual data (instructions, descriptions). Unlike a classic image generation model, which creates an image from text, a VLM can analyze an existing image and make modifications based on a prompt. For example, a user could ask Brush to "change the background color to blue and align the text to the right" on an already designed poster. This approach allows for a more natural interaction with design tools. Multimodal Transformer Architecture The transformer architecture is a type of neural network initially designed for natural language processing. Adapted for multimodal data (text + image), it combines an encoder, which analyzes the input data (in this case, an image and an instruction), and a decoder, which generates an output (a modified image). This structure allows the model to learn complex relationships between visual and textual elements. For example, if a user asks Brush to "remove the logo in the top left", the model identifies the relevant area and applies the modification without altering the rest of the design. Paid API An API (application programming interface) is a set of protocols that allow software to communicate with each other. In the case of Brush, the paid API means that developers or companies can integrate the model's features into their own tools for a cost per use (here, 0.005 USD per generated image). This economic model is common in generative AI: players like OpenAI (with DALL·E) or Stability AI (with Stable Diffusion) also charge based on usage. For Canva, this approach aims to monetize innovation while expanding its ecosystem. Anonymized Data Anonymized data is information in which elements that allow a person to be identified have been removed or modified. In the case of Brush, Canva indicates that it used user data to train its model, but without specifying the degree of anonymization. This practice, common in AI, raises ethical and legal questions, particularly regarding user consent and privacy protection. Regulations such as the GDPR in Europe strictly frame the use of such data. Analysis: what is at stake The launch of Brush by Canva is part of a broader trend toward the democratization of artificial intelligence (AI) tools applied to design, with major economic and competitive implications. From a mechanical standpoint, Canva's offering is based on a usage-based pricing model (0.005 USD per generated image), which could stimulate demand by reducing marginal costs for professionals. According to the company's internal tests, the latency of less than two seconds per request and the announced productivity gains (up to 69% faster editing) could encourage…