Pitch deck to video

The person brings a finished deck and says who pitches it. You deliver a 1920×1080 video, 1–3 min, of the presenter walking through the real slides: it should feel like the founder recording a walkthrough for one investor, not an ad. The slides are shown as they are; the presenter, the voice and the layouts are guide-present-with-an-avatar's. Rounds and deliverables follow the core skill.

Steps

1. Slides as images: 1920 px wide (a 4:3 deck → 1920×1440), source/slideNN.png, and a contact sheet of them (python-contact-sheet, images). Drop duplicates (an older copy at the end, a slide that restates another) and say which.
2. Who presents (guide-present-with-an-avatar, its first question): the presenter's name as said, an AI avatar or the person on camera, and any house rules (no added on-screen text: then the slides' words are the only ones on screen).
3. Script, one section per slide (guide-make-a-script-from-source, the deck route): speaker notes are the first draft; with none, the pitch arc: the problem, the solution, how it works, the market, traction, the team, the ask. About 2.5–3 words a second, spoken to "you", numbers as they are said. Gate: the script word for word, the presenter and the cost.
4. First look of the layouts (hn4a-compositor, The first look): the title slide and the first content slide in two or three presenter layouts, a still or a silhouette for the presenter. Gate: the layout mix.
5. Presenter, voice and the whole-track take (guide-present-with-an-avatar, Process: python-plan-lipsync-chunks, python-assemble-lipsync-chunks), or the person's own recording cut to its best takes. Its gates stand.
6. The cutout (python-matte-a-talking-head) of the assembled take, free.
7. The layout (hn4a-compositor-motion-graphics), one build script: the slides as they are at 1440×1080 on the left; a 480 px strip on the right, dark with a faint dot grid; slides change mid-gap with a 0.35 s zoom; the presenter layouts mixed, never the same twice in a row (guide-present-with-an-avatar, Layouts): the big cutout on the greeting and the close, the normal cutout, the box in the strip, the bubble. A product slide whose screen is the point moves (guide-show-a-product-on-screen). A still per layout, checked that nothing covers a slide's words. Gate: the live spec (hn4a-compositor, Live spec, then render).
8. Render and deliver: frames equal round(duration × 30); a contact sheet looked at; deliver the full file and a 1280 px copy (CRF 26).

Choices this workflow makes

• Real slides, never redrawn: their words stand in for captions, and the deck the investor gets later matches the video.
• One talking head, used every way: the box and bubble crop the footage, and only the cutout uses the matte.
• 4:3 slides at full height leave a 480 px strip for the presenter, so the slide is never covered.

Dead ends

• A big cutout over a full-size end card: it covered the tagline. Shrink the end card to 1280×960 on that layout.
• Lip-sync prompts from the transcript: misspelled names and "$2.99" for "two ninety-nine". Write them from the script.

Worked example

The shape of the demo clip: a pilot proposal deck's title slide (a two-line headline, a photo grid, "Prepared for" the prospect) on the left, the presenter, an AI avatar, cut out on the right over a light strip, speaking the opening line. On the same engine, a personalized 100.6 s version of a deck with an existing avatar and voice cost about $9.90 all in (workflow-sales-video-from-pitch-deck).