// AI NEWS · FEATURE

HeyGen Bets Avatar Video Belongs to AI Agents: Video Agent Now Plans, Renders, and Runs From a Prompt Inside Claude and ChatGPT

HeyGen has rebuilt avatar video around agents — its Video Agent turns a prompt into a reviewable plan before it renders, ships as an API you can call from Claude or ChatGPT, and sits alongside HyperFrames, its open-source HTML-to-video engine built so agents can render video from code.

KompozyTurn one idea into a week of content — across every platform, published for you.
Get Started →

2026-07-22 · by Moe Ameen

What happened

HeyGen, the identity-first AI avatar platform that said it crossed $200 million in ARR in June 2026, has spent 2026 reorganizing its product around a single idea: that AI agents, not human editors dragging clips on a timeline, will create most video. The clearest expression is Video Agent — a rebuilt creation layer where you describe the video you want and the agent returns a full "creative blueprint": a script, an avatar, a scene structure, pacing, suggested B-roll, narration, and captions. Instead of rendering immediately, HeyGen added a review-and-confirm step so you approve the plan first and refine it conversationally — shortening a scene, swapping the intro graphic, adjusting poses, lighting, or text overlays — before any credits are spent. On approval, the agent sources stock media from libraries such as Pexels, Getty, Unsplash, and Storyblocks, generates the voiceover, animates the avatar, applies Brand System caption styling, and renders the cut.

The second move is making that agent callable. In February 2026 HeyGen shipped a Video Agent API for prompt-to-video workflows and installable "Skills" that let an AI coding agent like Claude drive it: install the skill, add an API key, and tell the agent what you want — it optimizes the prompt, calls the API, and returns a finished MP4, so video becomes just another output of an agent task rather than something you build in a web editor. HeyGen paired this with a ChatGPT integration for describing a video and getting a produced result in the chat. Around the same launch it moved API usage to pay-as-you-go, ending free API credits.

Underneath sits a third bet on the same thesis: HyperFrames, HeyGen's open-source framework that turns plain HTML, CSS, and animations into deterministic MP4 video — the argument, from HeyGen's James Russo, being that large language models already speak HTML natively, so HTML is the right medium for agent-generated video. Together, Video Agent (a hosted prompt-to-video product) and HyperFrames (an HTML rendering primitive agents can operate themselves) are two halves of the same strategy: put video generation inside the agent workflow instead of a manual editor.

Why it matters for creators

  • Drafting a presenter video is collapsing from an afternoon in an editor to a single prompt plus a plan you approve — the bottleneck moves from making the video to deciding what to make.
  • The review-before-render step is the useful part: it keeps a human directing the cut and stops the agent from burning credits on a version you did not want.
  • An API-and-Skills model means video can be triggered from inside the tools creators already run — a Claude task, a ChatGPT chat — not just HeyGen's site.
  • Pay-as-you-go API pricing (no free credits) means agent-driven video is metered per render, so volume workflows need to budget by the minute.
  • None of this closes the last mile: an agent that produces one polished video still leaves captioning-per-feed, format fan-out, and cross-platform scheduling entirely to you.

How to act on this with Kompozy

The important detail for creators is where HeyGen's agent stops. Video Agent is a brilliant drafting machine — one prompt, one directed, avatar-led cut — but its output is a single MP4, and a single MP4 is not a content week. It does not reframe itself for six feeds, become a carousel and a thread and a blog and a newsletter, or schedule anything. HeyGen has automated the making of the video and left the operation around it untouched. That operation is exactly what Kompozy is.

You can act on this today from either end. If you want the whole thing in one place, Kompozy generates HeyGen-class avatar video natively — Persona HeyGen for multi-scene cuts, Persona Frames to composite the avatar into a brand-exact template — then fans that one idea into a Carousel, Quote Graphic, X thread, Blog Article, and Email Newsletter, all in one voice through your Persona Brief, and auto-captions, reframes, and publishes the set across Instagram, TikTok, YouTube, LinkedIn, X, Facebook, Pinterest, and Threads from a single queue. If you would rather draft in HeyGen's Video Agent, export the MP4 into Kompozy and let it run the repurposing and multi-platform publishing the agent skips. Better still, point Kompozy's autopilot at your sources so the whole cadence — generate, caption, schedule, publish — runs on a recurring basis. HeyGen made avatar video an agent output; Kompozy makes it a published, on-brand content operation.

Quick takeaways

  • Video Agent plans a video from a prompt, lets you approve and refine the blueprint, then renders an avatar-led cut.
  • It runs as an API and via Skills you can call from Claude, plus a ChatGPT integration; API usage is pay-as-you-go.
  • HyperFrames, HeyGen's open-source HTML-to-MP4 engine, is the same agent-first thesis one layer down.
  • The gap is distribution: one finished video is not captioned per feed, repurposed, or scheduled — Kompozy is the engine that does that.

Frequently asked questions

What is HeyGen Video Agent?

It is HeyGen's prompt-to-video AI agent. You describe a video and it returns a full plan — script, avatar, scenes, pacing, B-roll, narration, and captions — that you review and refine in chat before it renders. Once approved, it sources stock media, generates the voiceover, animates a HeyGen avatar, and builds the finished cut.

Can I run HeyGen Video Agent from Claude or ChatGPT?

Yes. HeyGen shipped a Video Agent API in February 2026 plus installable Skills so an agent like Claude can call it — install the skill, add an API key, and it returns a finished video from your prompt. HeyGen also has a ChatGPT integration for prompt-to-video in the chat. API usage moved to pay-as-you-go around that launch.

How is Video Agent different from HyperFrames?

Video Agent is a hosted product that plans and renders an avatar video from a prompt. HyperFrames is an open-source framework that turns HTML, CSS, and animations into deterministic MP4 video for agents to render themselves. One is the finished prompt-to-video product; the other is the code-first rendering primitive.

Does HeyGen Video Agent publish to social platforms?

No. It generates one finished video inside HeyGen. Reframing per platform, repurposing into other formats, and scheduling are not part of it. Kompozy generates HeyGen-class avatar video and then auto-captions, reframes, repurposes, and publishes it across nine platforms from one queue.

Related news

← All AI news · Get started →