AI writing has a shape
105.2k plays105,242 public plays. Primary reference for split framing, evidence zooms and phrase captions.
00:00–00:07 split with paper · 00:12–00:23 magnifier · 00:58–01:13 full camera
2 Oct 2026 · Reviewed 15 October 2026House style
Direct-to-camera teaching, short captions and evidence you can actually see. The familiar rhythm of Will’s recent strongest performers.
Will FrancisCurrent house style
The editing recipe
9:16 first · adapt to the footageLet the evidence carry the explanation.
In the reference The AI-writing reel opens immediately with the presenter above a research paper. Full-camera passages interrupt the split layout when the explanation becomes conversational.
- Open on the specific finding and its actual evidence. Avoid a separate title sequence.
- Trim mistakes, repeats and unhelpful gaps at phrase boundaries. Keep enough breath for a natural delivery.
- Use the split layout while showing a source, then return to full camera for interpretation or a caveat.
- Let a readable source view hold through a complete point. Do not impose a cut every two seconds.
Starting settings
Starting point: 1080 × 1920. Give the presenter roughly 40–50% of the height and the evidence the remaining space, adjusting to the source.
Keep the person natural and the studio recognisable.
In the reference Both references use a clear, contrast-rich studio image with blue/purple and magenta background separation. The apparent face temperature varies between clips.
- Correct exposure and white balance before shaping the look. Protect skin colour and highlights.
- Preserve the actual purple/blue studio lights; do not impose that tint on the face or screen capture.
- Keep document whites neutral and small UI text sharp. Grade the camera and evidence independently.
- Use modest contrast and saturation, matching adjacent shots by eye and scopes.
Starting settings
No extracted LUT or fixed temperature is available. Match the approved source visually; grading settings depend on the camera original.
Short, steady phrases at the join.
In the reference Bold white sentence-case captions sit on a dark rounded backing. They appear around the split line or over the lower presenter area, a few words at a time.
- Use Avenir Next Bold, as specified in the brand guide. Do not substitute an analysis tool’s guessed font.
- Start at 3–6 words per phrase and a maximum of two lines. Break at meaningful units of speech.
- Use a charcoal backing with consistent padding. Keep the position stable between phrases.
- During split scenes, put captions at the join without covering a face or relevant evidence. Check platform overlays on a real phone.
Starting settings
Suggested production start: 56–64 px at 1080 × 1920; white text on #272727. Tune to the actual crop. This is an adaptation setting, not an extracted source measurement.
Show the real thing, then focus the eye.
In the reference The AI-writing example uses actual paper figures, cursor movement and circular magnifiers. The Astra example combines genuine software captures with explanatory visual inserts.
- Use real product recordings for actions the viewer can repeat. Establish enough interface context before cropping in.
- Make one deliberate pan, zoom or magnifier move to the exact sentence or chart feature being discussed.
- Use the approved drawn explainer system only for abstract mechanisms that cannot be demonstrated directly.
- Keep source colours intact in charts. Use brand yellow for your own emphasis and one diagram per explanatory beat.
Starting settings
For new HTML explainers: 800 × 800, chunky CSS/SVG objects, 54 px minimum text, one purposeful explanatory animation, as required by the current skill.
A clear voice is the main track.
In the reference Hard cuts dominate. The research reel has no intrusive sound effects; the Astra analysis reports a quiet synth bed. A soundtrack is not the defining feature.
- Keep dialogue continuous across an evidence insert. Cut picture to meaning, not to a music beat.
- Use gentle noise cleanup, de-essing and compression only when needed. Preserve natural emphasis.
- Default to no music for dense research. If music helps, keep an unobtrusive instrumental bed well below speech.
- Use straight cuts between camera and evidence. Avoid decorative whooshes, zoom transitions and repeated sound effects.
Starting settings
Suggested delivery target: around −16 LUFS integrated, true peak at or below −1 dBTP. Check the whole programme and platform encoding; this is a production target, not a source measurement.
Use it for
Research breakdowns · AI explainers · Product updates
Use this style
Copy the brief into your editing chat with your script and footage. The skill includes the complete recipe and reference notes.
Read the editing brief
Use the Will Francis House style for this video. Read the attached style recipe and current brand guide. Build a speech-led edit from my footage and script, with natural skin tones, the existing studio colour, Avenir Next Bold phrase captions on charcoal, and large authentic evidence below the presenter. Use full camera for interpretation and straight cuts at changes of meaning. Use the current explainer-slideshow workflow for abstract diagrams. Bring me a shot-by-shot edit plan with source timecodes, caption breaks and any missing evidence before rendering. Do not invent results, footage or metrics.
The references
Based on the two highest-played reels in the 12-reel public sample returned on 15 October 2026. The sample covers 25 August to 2 October, not the whole account. Views are context, not proof that the edit caused performance.
GPT-6 Astra is coming
91.6k playsPublic play count rounded as returned. Secondary reference for studio treatment and alternating camera, software and diagram scenes.
00:02–00:42 evidence inserts · 00:42–00:59 camera · 00:59–01:12 software
4 Sep 2026 · Reviewed 15 October 2026