HyperFrames Topic-to-Video Setup on Windows (2026)
HyperFrames Topic-to-Video Setup on Windows (2026)
July 13, 2026
Keston CollinsVideo editor with nearly 10 years of experience, exploring the intersection of motion graphics and AI.
If you searched for "hyperframes topic2video setup," you want one thing: type a topic, get a finished narrated video, on a Windows machine. That workflow exists. But the name does not, and knowing that saves you an hour of confused searching.
One note before anything else. We have not run this pipeline on a Windows test bench ourselves. Everything below comes from the official HyperFrames documentation at hyperframes.heygen.com, the repo README at github.com/heygen-com/hyperframes, and community reports from people who have run it on Windows. Where a step comes from a community project rather than HeyGen's docs, we say so.
TL;DR
There is no command called topic2video in HyperFrames. The workflow people mean is served by two real routes: the official faceless-explainer skill, which turns any text or topic into a narrated explainer video, and a community Archon pipeline that automates the whole run from a single prompt. Both work on Windows. Both need Node.js 22+ and FFmpeg. The community route adds Python 3.10+, bun, jq, and espeak-ng for local Kokoro text to speech.
What "topic2video" actually maps to
HyperFrames is HeyGen's open source framework that renders video from HTML. You write a composition in HTML, CSS, and GSAP, and the CLI captures it frame by frame in a headless browser, then encodes the frames into an MP4 with FFmpeg. The repo ships 20 skills for AI coding agents, and we checked the full list: embedded-captions, faceless-explainer, general-video, motion-graphics, website-to-video, and the rest. Nothing named topic2video.
Two things answer the intent behind the search:
The official faceless-explainer skill (in the main repo). Give it a topic, an article, or rough notes, and an agent like Claude Code scripts, narrates, designs, and renders a full explainer video. Visuals are built in HTML per scene: typography, diagrams, data visuals. No stock footage capture.
The community Archon pipeline at github.com/coleam00/hyperframes-ai-video-generation. It chains research, scriptwriting, text to speech, and HyperFrames rendering into one command that outputs a vertical short of roughly 24 to 30 seconds.
Pick route 1 if you want control over each step. Pick route 2 if you want one command in, one video out.
Prerequisites for Windows
Here is everything the two routes need, based on the official README and the community repo's documentation.
Tool
Version
Needed for
What it does
Node.js
22 or newer
Both routes
Runs the HyperFrames CLI and framework
FFmpeg
Current
Both routes
Encodes captured frames into MP4
Headless Chrome
Bundled via Puppeteer
Both routes
Renders and captures each frame
Claude Code (or Cursor, Codex, Gemini CLI)
Current
Both routes
The agent that drives the skills
Python
3.10 or newer
Community route
Runs the Kokoro TTS scripts
espeak-ng
Current
Community route with Kokoro
Phoneme backend for Kokoro voices
bun and jq
Current
Community route
Workflow scripting and JSON handling
The official faceless-explainer skill handles narration through HeyGen's hosted TTS if you sign in, or falls back to Kokoro running locally if you do not. The community pipeline defaults to local Kokoro, with ElevenLabs as a paid alternative.
Two keys people expect but do not need: the official faceless-explainer skill requires no Pexels key, because it builds visuals in HTML instead of pulling stock footage, and no Gemini key, because your coding agent does the research and writing.
Route 1: The official faceless-explainer skill
This is the path documented by HeyGen itself.
Install Node.js 22+ and FFmpeg. On Windows, winget install OpenJS.NodeJS Gyan.FFmpeg handles both. Open a new terminal afterward so PATH updates take effect, then confirm with node --version and ffmpeg -version. If either command is not recognized, the install did not land on your PATH, which is the single most common Windows setup failure.
Install the HyperFrames skills. In your agent's terminal, run npx skills add heygen-com/hyperframes --full-depth. The official quickstart is specific about keeping the --full-depth flag: without it, the installer fetches a cached blob that lags behind the main branch by hours.
Check your sign-in status. Run npx hyperframes auth status. Signed in, narration uses HeyGen voices. Not signed in, the skill falls back to Kokoro locally. The skill's own workflow requires this check before anything renders, so do not skip it.
Run the skill with a topic. In Claude Code, invoke the faceless-explainer workflow and hand it your topic or source text. It locks a brief with you (length, aspect ratio, message), writes a storyboard and script, generates narration and word timings, builds each frame as HTML, and assembles the timeline.
Preview and render.npx hyperframes preview opens a live browser preview. npx hyperframes render produces the MP4. If the render completes, your whole toolchain is healthy.
If you have never installed HyperFrames at all, start with our How to Install HyperFrames guide first. This article assumes that base install works and focuses on the topic-to-video layer on top of it.
Route 2: The community Archon pipeline
The repo at github.com/coleam00/hyperframes-ai-video-generation wires the full run into one command. Its README documents Windows explicitly, which is rare and worth crediting. Steps below are from that README.
Install the system dependencies.winget install OpenJS.NodeJS Python.Python.3.12 Gyan.FFmpeg jqlang.jq, then install bun with irm bun.sh/install.ps1 | iex in PowerShell.
Install espeak-ng. Kokoro needs it as a phoneme backend. On Windows the README says to download the .msi installer from the espeak-ng GitHub releases page rather than trying command line installs.
Set up a Python virtual environment.python -m venv .venv, then activate it with .venv\Scripts\activate in PowerShell. Install the TTS packages: pip install kokoro python-dotenv soundfile numpy. The README warns you must activate this venv in every shell where you run the workflow.
Configure the environment file. Copy .env.example to .archon/.env and set KOKORO_VOICE=af_heart or any voice from the Kokoro-82M catalog on Hugging Face, plus KOKORO_LANG_CODE=a for American English.
Point the workflow at Claude Code. On Windows, set $env:CLAUDE_BIN_PATH = "$env:USERPROFILE\.local\bin\claude.exe" in your PowerShell profile. The README lists a missing or wrong path here as a common failure.
Run it.archon workflow run create-classic-short --no-worktree "Your topic here". The README quotes 20 to 30 minutes to a browser preview, then npx hyperframes render videos/<slug> for the final MP4.
Windows errors users have actually reported
These come from the project's GitHub issues and the community repo's known-issues list, not from our own testing.
Rendering fails with a HeadlessExperimental.beginFrame error. Reported in issue 294 on the HyperFrames repo: Chromium 147 and newer removed the beginFrame command that HyperFrames used for frame capture. HyperFrames 0.4.2 and later detect this and fall back to screenshot capture automatically. If you hit it, update HyperFrames, or install a compatible browser with npx puppeteer browsers install chrome-headless-shell.
Kokoro stalls on first run. The first narration triggers a model download of roughly 325 MB from Hugging Face into your user cache. On a slow or filtered network this looks like a hang. Let it finish once and every later run works offline, per the community repo's documentation.
Bun scripts truncate on Windows. The community repo documents that multi-line bun scripts cut off at the first newline on Windows, and it ships a bash wrapper workaround for the affected step. If you fork or modify the pipeline, keep that wrapper.
FFmpeg or Node not found after install. A PATH problem, not a broken install. Winget installs need a fresh terminal session before the new commands resolve. Verify with ffmpeg -version and node --version before blaming anything else.
Git LFS missing when cloning the full repo. Only relevant if you clone github.com/heygen-com/hyperframes for development rather than using npx. The README's Windows instruction is winget install GitHub.GitLFS, then git lfs install once per machine.
FAQ
Is there an official topic2video command in HyperFrames?
No. We checked the full skill list in the official repo and nothing carries that name. The faceless-explainer skill is the official workflow that turns a topic into a video, and community pipelines automate it further. Treat "topic2video" as a description of the goal, not a command to type.
Do I need a Pexels or Gemini API key?
Not for the official skill. Faceless-explainer builds its visuals in HTML per scene instead of pulling stock footage, and the research and writing happen inside your coding agent. The only account decision is narration: HeyGen sign-in, local Kokoro for free, or ElevenLabs on the community route.
Does this run on native Windows or do I need WSL?
Both the official docs and the community repo give native Windows commands through winget and PowerShell, so WSL is not required. The community route has more moving parts on native Windows (espeak-ng via .msi, the bun truncation quirk), so if you already live in WSL, that is a reasonable place to run it.
How long does one video take?
The community pipeline's README quotes 20 to 30 minutes from prompt to preview for a short of about 24 to 30 seconds. The official skill's duration depends on how much you collaborate on the brief and storyboard along the way.
Which route should I start with?
The official faceless-explainer skill. It has fewer dependencies on Windows, it is maintained in the main repo, and it teaches you the storyboard and render loop that every other pipeline builds on.
Two honest paths from here
If you want this pipeline, you want it for a real reason: full control of the composition, HTML as your video format, and renders you can regenerate from code forever. HyperFrames delivers that, and the setup above is the price of admission. For how it compares against the other major code-video framework, read HyperFrames vs Remotion.
If you read this far and the venv activation, PATH checks, and phoneme backends sound like a week you do not have, there is a different path that trades control for speed. A Motion Agent runs in the browser on a video creation platform: describe the clip, it fills a market-tested motion template, you export in minutes. No Node, no FFmpeg, no Python. With AutoAE that costs $9.9 a month or $2.9 per video. It will not give you a code-defined pipeline, and it is not trying to. Builders take the HyperFrames road. People with a Friday deadline take the other one. Both are legitimate.
