Sume is developing video generation APIs tailored for agents.

Sume co-founder David Im opens the waitlist for a video generation API layer designed for AI agents to create deterministic, publish-ready UGC content.

David Im, co-founder of Sume, is opening the waitlist for a video generation API layer designed not for human prompting, but to be called by AI agents. The company's bet: foundation video models are stochastic and their quality degrades as scenes stack up, making them ill-suited for automated production.

In response, Sume offers what it describes as deterministic, prop-based APIs—structured inputs rather than open-ended prompts—intended to return a publish-ready video on the first attempt, without wasting credits on iterations. The endpoints are intentionally "opinionated," with production-oriented default settings and guardrails, and designed to plug into existing agent stacks. The first announced component focuses on avatar and UGC (User-Generated Content) style, with the company promising to expand the API surface weekly towards cinematic, video understanding, or marketing content.

Access is via a waitlist, with a first API expected in early July. At this stage, Sume remains in pre-launch, without detailed public demonstrations, documentation, or pricing displayed.