Jake Moran joins Peter Yang for a 42-minute walkthrough of AI-assisted launch videos, talking-head edits and reusable visual styles in HyperFrames.

Give the agent editing language and accurate timing

  • Start with a rough cut: remove filler words, repeated phrases, long pauses and abandoned takes while preserving the story.
  • Use specific terms for zooms, cuts, B-roll and transitions, then describe the energy you want.
  • Moran uses local Parakeet transcription for precise word timestamps; bad timings can clip words even when the editing instruction is clear.
  • Keep music quiet beneath speech and listen to the result. Moran’s suggested volume setting is a starting preference, not a loudness standard.

Approve the story before rendering everything

  1. Provide the source assets, intended audience, rough story and brand references.
  2. Ask for a storyboard of still frames, then correct the sequence before the slower full build.
  3. Structure a feature launch around hook → context → payoff or call to action → brand.
  4. Inspect generated assets and the finished edit, including choices the agent made without being asked.
  • Moran demonstrates a background-removal launch assembled from three screenshots and additional generated assets.
  • The avatars come from HeyGen through a connected tool; Claude orchestrates the work rather than supplying every media model itself.

Build a reusable visual reference library

  • Supply a brand kit with actual colors, fonts and assets rather than asking vaguely for “my brand.”
  • A short reference clip or screenshot can communicate layering, text motion and pacing more clearly than a long description.
  • HyperFrames’ “moments” library supplies reusable scene references that can be adapted to the new story and brand.
  • Keep effects restrained in long tutorials so the speaker stays central; short social clips may need a different pace.

Make revisions cheap and specific

  • Draw on the relevant frame, attach a note and queue several changes before sending them to the agent.
  • Split longer videos into manageable segments, edit them separately and stitch the result together.
  • Yang describes a blur pass for sensitive names and emails. Review those frames before publishing.
  • Save a successful repeatable editing process as a skill or reference, including the steps that produced the result.
  • Caveat: this is a product demonstration. Transcription accuracy and time savings are personal reports, and generated likenesses need consent and review.

“You probably still need to know a rough storyline that you want to convey.” — Jake Moran, 16:21–16:24