news
Alibaba Cloud's Smart Video agent claims hour-long video from one prompt, with no quality bar attached
Announced at the 2026 Apsara Conference, the agent targets short-drama studios and advertisers, routes to multiple video models including Wan, and ships without published pricing, benchmarks or sample evaluation.
Alibaba Cloud used the third day of its 2026 Apsara Conference to announce Smart Video, an AI video agent that it says automatically produces videos up to one hour long from a single prompt, aimed specifically at short-drama studios and advertisers. The announcement gives a maximum duration, a language count and a target customer. It gives no pricing, no availability date outside the conference framing, and no evaluation of whether an hour of automatically assembled output is watchable.
What the agent is described as doing
Per Alibaba Cloud's release, a creator states an idea and the agent returns a finished cut, which can then be refined through further natural language instructions. The platform is model-agnostic in the sense that it supports multiple video models, including Alibaba's own Wan family, and it handles more than 20 languages. Smart Video was announced alongside Smart Studio, a system for standing up Model-as-a-Service infrastructure, as part of a broader set of AI capabilities and a global infrastructure expansion that includes first cloud regions in Turkiye, Finland and the Netherlands over the next 12 months.
Sixty minutes is a container size, not a quality claim
Duration is the easiest number to publish and the least informative one. A 60-minute ceiling tells you the agent will keep assembling shots until it hits an hour; it says nothing about character consistency across that hour, shot-to-shot continuity, how much of the runtime is usable without manual re-cutting, or how many generations it burns to get there. Short-drama production is exactly the use case where those failures compound, because the format depends on the same faces recurring across dozens of scenes. For reference on what the long-form end of this actually looks like when someone publishes the work, a studio recently shipped a 135-minute fully AI-generated feature, and the interesting part of that project was the production process, not the runtime.
HappyHorse shows up in a named customer deployment
The same release names Loomi Entertainment Group, a Malaysia-based entertainment company, as having integrated Alibaba's visual generation models through Model Studio into its own cinematic engine, specifically listing Wan and HappyHorse. That is a notable line, because HappyHorse is the model that arrived anonymously on the video leaderboard, took the top spot, and then lost it. Alibaba is now selling it under its own name as part of an enterprise content pipeline. The two companies say they are exploring Wan inside Loomi's micro-drama production workflow and have run two AIGC hackathons with more than 100 local creators trained.
What is missing before this is testable
There is no per-minute or per-generation cost in the announcement, no stated resolution or frame rate ceiling, no list of which third-party video models the agent can route to beyond Wan, and no sample output published with the release. Until Alibaba Cloud puts numbers or reference footage behind the hour-long figure, Smart Video is a product positioning statement aimed at a real and growing market, the Chinese and Southeast Asian short-drama business, rather than a capability anyone outside the conference hall can score.
Sources: Alibaba Cloud via Media OutReach, Alibaba Cloud, Unite.AI.
Priya Shenoy: Tracks what AI video actually costs across the platforms that resell access to the same handful of models. Treats a pricing page as a claim, not a fact, until someone checks it.