Canva Craft：Canva 第一个自研图像模型先规划再渲染——最高 2K、30 多种画幅、训练数据包含数百万自家模板，已向所有用户开放

2026 年 10 月 22 日（星期四，已按日历核算；日期由 TechCrunch 的署名行与 The Verge 的「according to an announcement on Thursday（据周四的公告）」携带——Canva 页面本身不带任何日期标记），Canva 发布「Introducing Canva Craft（Canva Craft 面世）」。按页面：「Today, we're unveiling Canva Craft, our first state-of-the-art image generation model, developed and trained by Canva Research, our in-house AI lab.（今天，我们发布 Canva Craft——我们的第一个先进图像生成模型，由我们的内部实验室 Canva Research 开发并训练。）」架构：「In the first stage, we designed an image planning model to interpret the prompt and write a detailed plan of the composition: which elements appear, their spatial relationships, relative scale, visual hierarchy, and any text that needs to appear in the image.（第一阶段，我们设计了一个图像规划模型来解读提示词、写出构图的详细计划：哪些元素出现、它们的空间关系、相对比例、视觉层级，以及图像中需要出现的任何文字。）」与「In the second stage, a dedicated image rendering model translates this plan into a high-fidelity image at up to 2K resolution across more than 30 aspect ratios.（第二阶段，一个专用图像渲染模型把该计划转译成高保真图像，最高 2K 分辨率、30 多种画幅。）」理由：「This allows the model to reason about the structure of an image before generating it, delivering greater precision, consistency and control over the final output.（这让模型在生成之前先推理图像的结构，为最终输出带来更大的精度、一致性与控制力。）」训练：「Its training data includes millions of human-made templates from Canva's content library, spanning professionally designed posters, social posts, presentations, infographics, menus and typography-rich layouts.（其训练数据包含来自 Canva 内容库的数百万个人工模板，覆盖专业设计的海报、社交媒体帖子、演示文稿、信息图、菜单与文字密集排版。）」——动词是「includes（包含）」——加上「Canva's professional designers worked alongside dozens of researchers to assess more than 25,000 pairs of images in head-to-head comparisons with leading models.（Canva 的职业设计师与数十位研究者一起，在与领先模型的两两对比中评估了超过 25,000 对图像。）」语言：「Prompts can be written in any language, and Craft can create text in 55 languages.（提示词可用任何语言书写，Craft 能用 55 种语言生成文字。）」厂商自己的 DesignArena 读数：「It debuted as the fourth most advanced image generation model on DesignArena, delivering industry-leading visual quality and accurate text rendering to support a wide range of creative tasks.（它在 DesignArena 上以第四先进的图像生成模型首次亮相，提供行业领先的视觉质量与准确的文字渲染。）」；「Craft outputs rank #1 spot globally for both People & Portraits (out of 87 models) and Landscapes & Nature (out of 85 models).（Craft 的输出在人物肖像（87 个模型中）与风景自然（85 个模型中）两个类别全球排名第一。）」；「It also ranks #2 for Abstract & Patterns and #4 for Products, placing it among the world's best models across multiple subcategories.（它在抽象与图案类别排第二、产品类别排第四，在多个子类别中位列世界最佳模型。）」；以及实验室说法「Built specifically for design, Craft makes Canva the world's second most advanced AI lab for image generation, as measured by rankings from DesignArena, the largest crowdsourced assessor of AI design.（专为设计而造，按 DesignArena——最大的 AI 设计众包评估方——的排名，Craft 使 Canva 成为全球第二先进的图像生成实验室。）」图层：「JSON boundary boxing makes it easier than ever to convert images into editable layers with Magic Layers, with more accurate and faster results.（JSON 边界框让把图像转成 Magic Layers 可编辑图层比以往更容易，结果更准、更快。）」安全：「Like all AI products on Canva, Craft is backed by Canva Shield, our comprehensive approach to AI safety.（与 Canva 上所有 AI 产品一样，Craft 由 Canva Shield——我们对 AI 安全的全面方案——支撑。）」可用性：「Canva Craft is available starting today to all Canva users globally.（Canva Craft 自今日起向全球所有 Canva 用户开放。）」，入口是 Dream Lab；一句话写「Craft is also available via Public API and the Canva MCP Server…（Craft 也可通过公共 API 与 Canva MCP Server 使用……）」而功能清单单独标注「Canva MCP (Coming Soon)（即将推出）」——卡片按页面自己的标注把 MCP Server 写成即将推出；以及「These advances underpin Craft-fast and Craft-ultrafast, two upcoming variants designed to deliver high-quality image generation at a fraction of the cost of leading frontier models.（这些进展支撑着 Craft-fast 与 Craft-ultrafast——两个即将推出的变体，以领先前沿模型零头级的成本提供高质量图像生成。）」——未给出任何上线日期。按 TechCrunch（Ivan Mehta，10 月 22 日下午 12:00 PDT）：「The company noted that rather than using photographs, it used designs to train this model using human-designed templates from Canva's library, including posters, social posts, presentations, infographics, menus, and typography-rich layouts.（公司指出，训练这个模型用的是设计而非照片，素材是 Canva 库里的人工设计模板，包括海报、社交媒体帖子、演示文稿、信息图、菜单与文字密集排版。）」——「rather than using photographs（用设计而非照片）」是 TechCrunch 的转写，Canva 页面的动词是「includes（包含）」；TechCrunch 还带着页面上没有的 Leonardo 层（「In 2024, Canva acquired media generation platform Leonardo.ai. It said that the researchers from Leonardo contributed to the new image model.（2024 年 Canva 收购了媒体生成平台 Leonardo.ai。公司表示 Leonardo 的研究者参与了新图像模型。）」），并把两个变体写成「Craftfast and Craft-ultrafast, which will be released soon（Craftfast 与 Craft-ultrafast，即将发布）」。按 The Verge（Emma Roth，10 月 22 日）：「The model is currently ranked fourth on Design Arena's leaderboard for the best AI image generation models, a crowdsourced benchmark for AI designs. It sits behind OpenAI's GPT-Image-2.5 Sunburst, GPT-Image-2.5 Flare, and GPT-Image-2.（该模型目前在 Design Arena 的最佳 AI 图像生成模型排行榜上排第四——一个针对 AI 设计的众包基准。它排在 OpenAI 的 GPT-Image-2.5 Sunburst、GPT-Image-2.5 Flare 与 GPT-Image-2 之后。）」