上海初创公司Sand.ai开发的MAGI-2 Preview在Artificial Analysis图生视频开源权重模型榜单上排名第二，其付费工具VidMuse允许用户选择MiniMax、字节跳动、快手和Google的模型生成视频。
VidMuse的Anime Studio使用字节Seedance 2.5可制作约30秒带对话和声音的动漫短剧，同时支持音乐视频、产品广告等用途。创始人曹越表示，即使客户使用竞品模型，产生的反馈也能帮助改进Sand.ai自有模型。
根据未审计数据，Sand.ai 2026上半年营收260万美元，净亏损2250万美元。上市公司网宿科技拟以3亿元人民币投资约4.4%股权，投后估值约10亿美元，协议尚未签署。
文章认为这一策略既能通过实用工具赚钱，又能收集多模型用户反馈，但风险在于初期需求能否转化为足够复购以支撑实验室运营。DataEye数据显示2026上半年抖音新增AI和动画短剧22.19万部，美国IAB报告显示近三分之二数字视频广告买家已在使用生成式AI。
文章分析了Sand.ai通过VidMuse工具聚合多家模型实现营收，同时用用户反馈迭代自有MAGI-2模型的策略，并披露了其上半年财务数据与潜在投资估值。
A Shanghai startup founded in 2024 ranks second among open-weight models on Artificial Analysis’s image-to-video leaderboard. Its paid video-creation tool also lets customers use rival models.
@sandai_hq develops MAGI-2 Preview, which generates video with sound from a text prompt, or a prompt plus a still image. It also runs VidMuse, which Chinese business outlet @YicaiChina identified in June as its core revenue product.
VidMuse offers tools for music videos, product ads and anime dramas. Customers can choose models from Chinese developers @MiniMax_AI, @BytedanceTalk and Kuaishou, as well as @Google. According to its product guide, VidMuse’s Anime Studio uses ByteDance’s Seedance 2.5 to assemble roughly 30-second anime dramas with dialogue and sound.
Founder Cao Yue says feedback from people using VidMuse could help improve http://Sand.ai’s own models, even when users choose another company’s model.
http://Sand.ai recorded $2.6 million in revenue and a $22.5 million net loss in the first half of 2026, according to unaudited figures disclosed by a prospective investor.
That investor is Wangsu (Wangsu Science & Technology), a listed content-delivery provider proposing to invest RMB300 million ($44.5 million) for roughly 4.4% of http://Sand.ai in its A+ round. The terms imply a valuation of about $1 billion if all investors in the round complete their investments. The agreements were unsigned when Wangsu announced the plan on September 29.
Wangsu offers short-drama studios a gateway to use different models through one interface, and is expanding into model hosting and computing services. More video production could bring it revenue from the computing used to generate footage and from delivering it to viewers.
Research firm DataEye counted 221,900 new AI and animated short dramas on China’s Douyin video platform in the first half of 2026. In the US, the @iab’s July report found nearly two-thirds of digital-video ad buyers used generative AI for creative, up from half in 2025.
The strategy makes sense: earn from a useful creation tool while gathering feedback from users of several models. The risk is that early demand does not become enough repeat business to support the lab.
Full analysis on Substack:
https://research.zengenlabs.com/p/the-video-ai-company-making-money
来源：Dermot McGrath（麦德蒙） · x.com