设计平台Canva推出了名为Canva Craft的首个图像生成模型，以与Google、OpenAI、ByteDance和Meta等科技巨头的同类产品竞争1。该模型采用两阶段方法理解用户提示，可生成最高2K分辨率的图像，支持30种不同宽高比1。
Canva Craft的独特之处在于其训练方式1。与主要依赖照片数据的同行不同，该模型使用Canva库中的人类设计模板进行训练，包括海报、社交帖子、演示文稿、信息图表、菜单和排版丰富的布局1。这样的设计使其能够与Canva的Magic工具结合，生成分层设计1。2024年，Canva收购了媒体生成平台Leonardo.ai，其研究人员为新图像模型的开发做出了贡献1。
Canva计划通过API和第三方聊天机器人（包括ChatGPT和Gemini）提供该模型1。此外，公司即将发布两个成本效益版本：Craftfast和Craft-ultrafast1。由于2026年增长预测下调，Canva的估值从42亿美元跌至35亿美元1。
Design platform Canva has unveiled Canva Craft, its inaugural image generation model designed to compete with similar offerings from Google, OpenAI, ByteDance, and Meta 1. The model employs a two-stage approach to interpret user prompts and produce images at resolutions up to 2K, supporting 30 different aspect ratios 1. Unlike competing models trained on photographic datasets, Canva Craft was trained using human-created design templates from Canva's library, including posters, social media posts, presentations, infographics, menus, and typography-rich layouts 1.
The new model integrates with Canva's existing Magic tools to generate layered designs and will be made available through an API as well as third-party chatbots including ChatGPT and Gemini 1. Canva acquired media generation platform Leonardo.ai in 2024, and researchers from that platform contributed to the development of the new image model 1. The company plans to release two cost-optimized versions—Craft-fast and Craft-ultrafast 1.
The launch occurs against a backdrop of shifting company valuations. Canva is tracking 2024 revenue toward $4 billion, yet its valuation has declined from $42 billion to $35 billion due to downwardly revised growth projections for 2026 1.
评论
还没有评论，欢迎留下第一条。