Canva 发布首个自研图像模型 Craft
Canva launches Craft, its first in-house image model
AI 导读
设计平台 Canva 发布首个自研图像生成模型 Canva Craft，由内部实验室 Canva Research 从零训练，主要使用数百万份设计模板而非照片作为训练素材。
摘要与补充说明
Canva 称该模型可对标 Google、OpenAI、ByteDance 和 Meta 的同类模型。Craft 先生成图像构图、比例和文字，再输出最高 2K 分辨率、覆盖 30 多种宽高比的图像，并支持 55 种语言的文字渲染；用户可通过 Magic Layers 将结果拆分为可编辑图层。据 Canva 数据，Craft 在 DesignArena 排名中综合第四，人像与风景类别第一。
正文
Design platform Canva has released Canva Craft, its first in-house image generation model. The company says it can compete with comparable models from Google, OpenAI, ByteDance, and Meta. According to Canva, its internal lab Canva Research trained the model from scratch, relying mainly on millions of design templates rather than photos.
Craft first plans out image composition, proportions, and text, then generates images at up to 2K resolution across more than 30 aspect ratios. The model can render text in 55 languages, Canva says. Users can break results into editable layers using Magic Layers. On the DesignArena platform's ranking, Craft placed fourth overall and first in portraits and landscapes, according to Canva.
Craft is available to all users through Dream Lab and the public API, with MCP integration for chatbots like ChatGPT and Gemini planned for later. Beyond lower costs, the in-house model gives Canva more control over quality and speed. Researchers from Leonardo.ai, which Canva acquired in 2024, also contributed to the project. Two cheaper variants called Craft-fast and Craft-ultrafast are expected soon.
AI News Without the Hype – Curated by Humans
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.