Marketing videos

yottacode can help turn raw screen recordings and local assets into publishable marketing cuts by driving local ffmpeg/ffprobe tools through the agent. The recording workflow targets yottacode demos: inspect the recording, find dead air or visually idle terminal stretches, propose cuts, then render YouTube/X-ready MP4s and short GIF previews after approval. Prompt mode can also draft and render asset-based release videos from docs, logos, screenshots, and supplied clips.

Dependencies

Install the external binaries yourself; yottacode detects them but does not bundle or auto-install them.

Check readiness with:

yottacode doctor

The doctor output includes a Media Editing section with install hints for missing binaries.

Slash-command shortcut

Use /video in the TUI as a guided entry point:

/video
/video out/demo.mp4
/video edit out/demo.mp4
/video analyze out/demo.mp4
/video prompt Create a 45-second release video from RELEASE_v0.3.0.md and marketing/raw/demo.mp4

Bare /video explains the available video capabilities. With a path, it sends a normal agent workflow prompt that uses the same public tools (media_probe, media_analyze, media_render) and the same approval gates. With prompt <goal>, it asks the agent to plan an asset-based marketing video from local docs, screenshots, title cards, and clips, then stop for approval before rendering with media_compose / media_render. Natural language remains supported; /video is only a shortcut for discoverability. See video-tools.md for the full tool/capability matrix.

Prompt-driven marketing videos

Use /video prompt <goal> when you want a finished marketing-video plan rather than cleanup for one recording:

/video prompt Create a 60-second v0.4.0 release video from CHANGELOG.md, RELEASE_v0.4.0.md, and marketing/raw/dispatch-demo.mp4
/video prompt Make a 20-second X teaser from marketing/raw/demo.mov highlighting subagents and worktrees
/video prompt Create a YouTube tutorial intro for plan mode using screenshots in marketing/assets/

Prompt mode is still asset-based. The model reads referenced docs and assets, writes a storyboard/script, proposes segment order and output profiles, and asks for approval before calling any render tool. After approval it can use media_compose to assemble title cards, screenshots/images, and approved clip segments into a draft MP4 with branded templates, lower-third captions, simple fades, and image zoom/pan motion, then use media_render for platform exports.

This does not require an AI video model. Hosted text-to-video, generated b-roll, TTS narration, and music generation are optional future layers; the current workflow uses a language model for planning/copy and local ffmpeg for deterministic assembly.

Recommended workflow

1. Record the raw demo with your normal screen recorder.

2. Put the source file under the project, for example marketing/raw/demo.mp4.

3. Ask yottacode to inspect it:

   Probe marketing/raw/demo.mp4 and tell me the duration, resolution, and audio setup.

4. Ask it to find fluff/dead air:

   Analyze marketing/raw/demo.mp4 for terminal demo idle time and propose cuts. Use mode=auto. Do not render until I approve the edit plan.

5. Review the proposed cut list. Ask for changes if a pause should stay.

6. Render the final outputs:

   Render the approved edit plan to YouTube 16:9, X 16:9, and GIF preview outputs under marketing/out/.

For silent terminal recordings, media_analyze automatically falls back to visual idle detection. You can force the terminal-tuned thresholds with:

Analyze marketing/raw/demo.mp4 with media_analyze mode=terminal_demo and show candidate cuts; do not render yet.

Output profiles

When multiple profiles are requested from one base output, yottacode appends the profile name to the file stem, for example demo-youtube_16x9.mp4, demo-x_16x9.mp4, and demo-gif_preview.gif. GIF output uses ffmpeg palette generation for better quality, but GIFs grow quickly; use gif_preview or gif_preview_large for short approved clips rather than full-length demos. Ask for speed=1.5 or speed=2 when a terminal GIF should move faster without changing the approved cut ranges.

Captions and branding

media_compose can build title-card/image/clip drafts from local assets with branded templates, lower thirds, image zoom/pan motion, and simple fades. media_render accepts an optional captions file path and burns subtitles into the output with ffmpeg. Keep brand assets in the repo, for example under marketing/assets/, so normal path validation and review apply.

Current boundaries

This is not a full timeline editor and it does not generate AI video clips. It edits and composes real local assets with ffmpeg utilities. Generated feature videos, TTS narration, and stylized AI clips remain separate roadmap work in roadmap/video-generation.md.