HeyGen's companion mode lets coding agents direct video creation
The avatar-video company's new mode has AI agents pitch creative angles and share storyboards mid-flight, shifting generation from one-shot orders to direction.
HeyGen introduced companion mode on July 22, a new way for AI agents to make videos that replaces one-shot generation with an iterative directing loop. Announced in a thread on X, the feature lets an agent pitch multiple creative angles for a brief, check in with a storyboard before rendering, and refine the piece through review rounds — what the company characterizes as the difference between ordering a video and directing one.
Built for the agent ecosystem, not the web app
Companion mode lives in HeyGen's open agent tooling rather than its consumer interface. It extends the company's HyperFrames system — an HTML-based video composition layer built explicitly for agents — and its agent-skills repository, which plugs avatar creation, video generation and translation into coding agents such as Claude Code, Codex and Cursor. Recent additions include an intent layer, a review loop, and persistent user memory via a BRIEF.md file, so an agent can remember a creator's brand, tone and preferences across projects. The same day, HeyGen promoted its audio-to-video API: a voice recording plus a single image yields a talking video in one call, batchable to a hundred narrations over the same image.
The release lands amid a broader race to make generative video legible to agents, with Kling publishing MCP workflow tutorials the same week.
Why it matters: generative video's core adoption problem is control — one-shot prompting produces impressive but unsteerable output. Structured checkpoints like storyboard reviews attack that directly, and putting them inside coding agents rather than a web UI is a bet on where creative work is migrating. AIGC vendors are repositioning as infrastructure for the agent ecosystem: the customer is no longer only a human at a timeline editor, but an autonomous system with a brief, a memory file, and a budget.