Turn podcast episodes into captioned vertical clips

Oct 31, 2026

A podcast clip needs to make sense to someone who has never met the host. Choose a passage with a complete thought, include the context it needs and let the speaker finish. Framefields can turn those editorial decisions into a consistent vertical video layout.

The source format has a substantial audience: YouTube reported more than one billion monthly podcast viewers in May 2026. That does not guarantee demand for any individual clip, but it makes podcast repurposing a useful workflow to learn.

Choose the excerpt before the layout

Read a timestamped transcript alongside the recording. Pick a 35-second passage where a guest explains one mistake and how they fixed it. Keep the exact source in-point and out-point in your notes. If the answer relies on the host’s question, include it or provide a faithful context line.

Time	Layout	Editorial role
0–4 s	Speaker and short headline	Orient a new viewer
4–27 s	Speaker with captions	Deliver the complete idea
27–32 s	Same layout	Preserve the conclusion
32–35 s	Small episode reference	Point to the full conversation

Avoid opening on a dramatic sentence if the rest of the clip changes its meaning.

Assemble the clip

Use Layer.video with trimStartSec for the source in-point, then set the excerpt duration. Crop deliberately for a single speaker. When two people speak, use a designed split layout or planned cuts; do not assume automatic speaker switching.

Prepare an SRT whose timestamps start at the beginning of the exported clip. Subtract the source in-point from episode timestamps, remove cues outside the excerpt and proofread names. Use Layer.caption to display it. See media and captions for supported layers.

Keep the subtitle area away from faces and platform controls. Use short phrases that preserve meaning, and review the most text-heavy frame on a phone. Music is optional; clear dialogue matters more than filling silence.

Use Codex, ChatGPT, Claude, Cursor or Grok Bot

You can use Codex, ChatGPT, Claude, Cursor, or Grok Bot to plan the edit and write Framefields TypeScript. In a coding environment with project and terminal access, ask the assistant to implement the brief and run the checks. In a chat-only environment, ask for the source files, save them locally and run them in your Framefields project. Rendering needs Node.js 22+ and a supported WebGPU environment; see getting started.

For Codex or Cursor, install the Framefields skills:

npx skills add gatewai-dev/framefields -a codex -a cursor

For Claude Code, use /plugin install framefields. With ChatGPT or Grok Bot, supply the agent reference, your brief and asset list; available execution tools depend on your environment. The workflow does not require a native integration in every assistant.

Create a project with npx framefields init my-video --install --pm npm, then work inside my-video. Use the generated font setup and put your source media under assets/. Run npm run typecheck, inspect comp.lint() findings, render selected frames and a grid, then run npm run preview before npm run render. The agent workflow covers inspection and QA. Check portrait layouts against the destination app’s current controls before uploading.

Ask your assistant to edit

Build a 35-second Framefields podcast clip at 1080x1920, 30 fps.
Use assets/footage/episode.mp4 from the approved in-point in
assets/clip.json. Use the clip-relative captions in assets/captions/.
Preserve the speaker's complete argument. Use deliberate crops;
show both speakers when the notes request a split layout.
Add the approved headline and episode reference. No invented quotes,
automatic speaker switching or generated transcript assumptions.
Inspect subtitle-heavy frames and show a preview with dialogue.

Make the next excerpt easier

Store each clip’s source timestamps, title, speaker names, caption path and output ID together. A shared layout then supports a series without hiding which editorial decisions produced each clip.

Check the first and last words against the original episode. Also listen for clipped consonants at the cut and watch for captions ending early. A contact sheet catches layout issues; only playback catches whether a sentence has room to land.

For a clip built around music instead of speech, try audio-reactive visualizers.