talking_head_production v1.0.0

name: talking-head-production description: > Plans, shoots, and rough-cuts a talking-head video end to end: framing, eye-line, three-point lighting, teleprompter delivery, and jump-cut editing that removes dead air and filler words. Use whenever someone needs to film themselves or an interviewee speaking to camera, set up a webcam or home studio for recording, clean up raw talking-head footage, or is setting up lights and camera for an interview, explainer, or testimonial.

Talking Head Production

Purpose

Take a talking-head video from empty room to finished first cut: set up the shot, light the face, get a natural delivery, and edit out the dead weight. This covers both sides of the camera, the person filming and the person talking. Read references/shot-checklist.md on set. Run bin/footage-check.py on raw footage before editing.

Workflow

1. Set the shot

• Camera at eye level, on a tripod. Phone or camera is fine, 1080p minimum.
• Frame head and shoulders, medium close-up. Eyes sit on the upper third line of the frame, small gap of headroom above the head, slightly off-center toward the rule of thirds. Direct-to-camera: center is fine and feels conversational.
• Eye-line: lens-level for direct address. Interview: place the interviewer right beside the camera so the eyeline stays tight to the lens.
• Pull the subject 3 to 6 feet off the background for depth. Simplify what is behind them, cluttered backgrounds read as noise.
• Two-camera option: second camera at a slight side angle. Use it every 5 to 10 seconds to break visual monotony, and to hide awkward jump cuts later.

2. Light the face

Three-point setup, soft sources (window light counts):

• Key: main source, 45 degrees to one side of the face, slightly above eye level. Softbox, lamp bounced off a wall, or a window with a curtain.
• Fill: opposite side from the key, softer and dimmer, roughly half the key's strength, to open up shadows without flattening them. A white foam board works.
• Back/hair light: behind and above the subject, aimed at head and shoulders, subtle. Its only job is separating the person from the background.
• Keep color temperature consistent across all sources. Window daylight is around 5600K; match practical lamps to it or the white balance drifts.
• Budget ladder: no lights yet? Face a window for the key, hold white foam board as fill, any desk lamp with a sheet over it as back. One soft window beats three harsh bulbs.

3. Get the delivery

• Write how people talk: short sentences, 3 to 5 beats per section, not essay prose. Memorize beats, not words.
• Teleprompter: keep the script tight against the lens, short lines, clear paragraph breaks. Set scroll speed to match speaking pace so the eyes barely move. The person reads the lens, not the screen.
• Camera energy: speak louder and slightly more energetic than feels normal, it flattens out on playback. Stand if possible. Hands visible, gestures in frame.
• Record in short sections, 5 to 10 minutes each. Do the important material first while energy is high. Water nearby, room temperature.

4. Jump-cut the rough cut

• First pass is purely subtractive: long pauses, ums and uhs, false starts, repeated takes, production asides ("let me start that again").
• Cut rules that keep it sounding human:
• Never cut mid-word. Cut on silence, not on a breath.
• Leave a breath of air around cuts, about a quarter second, so joins do not sound clipped.
• Repeated takes: keep the best complete delivery, cut the rest. Do not Frankenstein a sentence from five takes unless there is no other option.
• Preserve the hook: do not restructure the first 30 seconds without the owner's say-so.
• Second angle or b-roll covers the worst jump cuts and resets attention every 5 to 10 seconds. A punch-in (scaling the same shot to 110 percent) does the same job with one camera.
• Listen through at normal speed before calling it done. If a cut sounds rushed or incomplete, open it back up.

5. Finish

• Captions: burn them in or ship an SRT. Most talking-head viewing happens muted on mobile.
• Audio: light noise reduction, consistent level across the cut, music bed low enough that it never fights the voice.
• Export: H.264, 1080p, match the source frame rate.

Output Contract

A finished first cut of the talking-head video, plus:

• The raw footage untouched (never destructive-edit the source).
• A cut list: timecodes of what was removed and why, so the owner can restore anything.
• Captions file alongside the export.

Operating Rules

1. Audio outranks video. Viewers forgive a grainy shot, nobody forgives unintelligible speech. Mic check before every take.
2. Never invent specs. Frame rates, light positions, and ratios here are starting points to adjust by eye, not laws.
3. Do not restructure someone's message in the edit. Tighten the delivery, keep the meaning. Content cuts get flagged to the owner.
4. One soft key beats a flat wall of light. If the face looks shadowless and gray, kill the fill, do not add more light.
5. Every factual claim about equipment or technique must trace to a primary source. Unverifiable tips get cut, not shipped.