Topic → Publish-Ready Long Video
The talking-head long-form pipeline: source-backed research, a structured script, a text-based edit, studio-clean audio and packaging that earns the click.
1–2 / week
Total time
180 min
Total cost
$72 per run, estimated
The pipeline
- 1
Source-backed research
30 minInput: Topic + your top sources
Output: One-page research brief with numbered sources
Upload the 5–8 sources you will actually cite and generate a source-backed brief — grounding beats general knowledge for anything with numbers or claims in it. Every claim that survives to the script keeps its source number; anything without one is opinion, and gets labelled as such.
Prompt
From these sources, produce a one-page research brief: the 5 strongest claims (each with its source number), 2 tensions or disagreements between sources, and 3 concrete examples or numbers I can show on screen. Nothing from your own knowledge unless marked [UNVERIFIED].
- 2
Script & outline
40 minInput: Research brief
Output: Numbered-section script with visual notes
Structure before prose: the section order is the video. Each section gets a claim, a proof, and a forward pull. Mark visual notes [VISUAL: ...] inline so the edit step has a shot list, not a guess. Read the opening aloud — if you would skip it, rewrite it.
Prompt
Turn this research brief into a video script. Structure: cold open (10s, strongest claim), context (what the viewer should care about), 4–6 sections each with claim → proof → forward pull, and an end that points to the next logical video. Number every sourced claim with its source number. Mark [VISUAL] notes where a screen or prop should appear.
- 3
Record & text-edit
60 minInput: Recorded takes + script
Output: Cut video, transcript-edited
Record against the script, not in one take. Edit the transcript: delete silences and stumbles like text, mild filler removal only (a heavy pass sounds like a chatbot). Keep the [VISUAL] moments in the cut — they are the edit plan.
- 4
Audio polish
15 minInput: Cut video audio
Output: Studio-clean voice track
Studio Sound (or equivalent) on the voice track; if one mispronounced word survived the edit, overdub just that line in your cloned voice instead of re-recording the segment. Music low and consistent under the voice, never under the key claims.
- 5
Thumbnail & packaging
20 minInput: Final cut + title options
Output: Thumbnail + title + chapters
Three thumbnail concepts tested at 120px wide. Chapters on (they change skip behaviour on long videos) and end screens that point to the next logical video. The title is a promise the first 30 seconds must keep.
- 6
SEO & publish
15 minInput: Video + packaging
Output: Published video with researched metadata
Research the search intent before writing the description. First 120 characters of the description are the visible ones — front-load the promise. Pin the comment with the one thing you want viewers to do next.
Keep in mind
- The research step is the moat: a video where every number has a source survives fact-check comments; a video where they do not, does not.
- Text-editing a 25-minute video is faster and better than timeline editing — but only if the transcript is fixed first (names, jargon, acronyms).
- 1080p is the standard for talking-head; save 4K for client work. Upload tax is real.
Total time
180 min
Total cost
$72 per run, estimated
The pipeline
- 1
Source-backed research
30 minInput: Topic + your top sources
Output: One-page research brief with numbered sources
Upload the 5–8 sources you will actually cite and generate a source-backed brief — grounding beats general knowledge for anything with numbers or claims in it. Every claim that survives to the script keeps its source number; anything without one is opinion, and gets labelled as such.
Prompt
From these sources, produce a one-page research brief: the 5 strongest claims (each with its source number), 2 tensions or disagreements between sources, and 3 concrete examples or numbers I can show on screen. Nothing from your own knowledge unless marked [UNVERIFIED].
- 2
Script & outline
40 minInput: Research brief
Output: Numbered-section script with visual notes
Structure before prose: the section order is the video. Each section gets a claim, a proof, and a forward pull. Mark visual notes [VISUAL: ...] inline so the edit step has a shot list, not a guess. Read the opening aloud — if you would skip it, rewrite it.
Prompt
Turn this research brief into a video script. Structure: cold open (10s, strongest claim), context (what the viewer should care about), 4–6 sections each with claim → proof → forward pull, and an end that points to the next logical video. Number every sourced claim with its source number. Mark [VISUAL] notes where a screen or prop should appear.
- 3
Record & text-edit
60 minInput: Recorded takes + script
Output: Cut video, transcript-edited
Record against the script, not in one take. Edit the transcript: delete silences and stumbles like text, mild filler removal only (a heavy pass sounds like a chatbot). Keep the [VISUAL] moments in the cut — they are the edit plan.
- 4
Audio polish
15 minInput: Cut video audio
Output: Studio-clean voice track
Studio Sound (or equivalent) on the voice track; if one mispronounced word survived the edit, overdub just that line in your cloned voice instead of re-recording the segment. Music low and consistent under the voice, never under the key claims.
- 5
Thumbnail & packaging
20 minInput: Final cut + title options
Output: Thumbnail + title + chapters
Three thumbnail concepts tested at 120px wide. Chapters on (they change skip behaviour on long videos) and end screens that point to the next logical video. The title is a promise the first 30 seconds must keep.
- 6
SEO & publish
15 minInput: Video + packaging
Output: Published video with researched metadata
Research the search intent before writing the description. First 120 characters of the description are the visible ones — front-load the promise. Pin the comment with the one thing you want viewers to do next.
Workflow builder
Questions & tips
No questions yet — be the first to ask.
Loading community posts…