Course promo
Hyperframes adds keyframes: AI video agents with a Hermes video agent
Julian Goldie demonstrates the open-source Hyperframes skill — now with keyframe and arc-motion support, so an agent can watch its own generated motion and self-correct it — paired with a Hermes-based video agent that researches a topic, writes a script, generates B-roll and an avatar presenter, and assembles a finished video from a single prompt. He also shows companion pieces (Hermes Astros for trending-topic research and Open Montage for cinematic clips) running inside his "Agent OS." Much of the video is a walkthrough of his paid community setup rather than a reproducible build, so treat it as a product tour.
Watch on YouTube → · ← Back to News
What you can actually set up from this
Extracted from the video's own transcript — the specifics the original summary left out.
Reproducible steps
- What changed
Hyperframes, a free open-source skill that lets an agent build videos by writing HTML, added keyframes and arc motion. The speaker says an agent using it can now see the motion in its own output, spot a problem and correct it in a loop.
- The video-agent flow shown (in the speaker's own Agent OS)
Pick a topic (surfaced by a keyword-monitoring setup called Hermes Astros) → set the length → choose presenter or audio-only → choose the B-roll generator (Grok or MiniMax were the two named) and the avatar → "write the script" (the agent drafts a hook and three scenes plus B-roll ideas) → approve or rewrite the script → "generate avatar B-roll" → assemble.
- Build it yourself
The speaker's own advice: give Claude the Hyperframes skill and ask it to build a "mission control" for a video agent step by step; Claude or Hermes Agent can do the building. Custom workflows (explainer, slideshow) can be added as presets.
- Another option
Open Montage, an open-source project the speaker uses for more cinematic AI-generated video.
Gotchas
- Course promo: the setup shown is the speaker's paid community product, and no commands or install steps are given in the video.
- The speaker recommends keeping a human check on the script and the footage before the final assembly.
- Claims such as AI videos often outperforming human ones are the speaker's observation, not a measurement.





