How to write a standing edit spec your channel reuses every episode

Last updated 2026-08-31 · 8 min read

What is the short version?

If you publish every week you are not briefing a video, you are running a channel with a standard. Re-explaining that standard from scratch each time is both tedious and lossy, and the fix is to write it down once and hand the same file over with every episode.

1. Write your channel's rules into one document.
2. Go to Vizard Agent, upload the raw footage and attach the file.
3. Say: follow this, do not reinvent, and report anything you changed.

What do you need before you start?

The spec, and honesty about what is actually a rule rather than a preference. Vizard Agent applies what is written, so a document that mixes hard constraints with vague aspirations produces exactly the inconsistency you wrote it to prevent.

• The tone. In your words, with what it is not.
• The non-negotiables. Numbered, unambiguous.
• The known quirks. Your footage's specific problems.
• The house look. Fonts, colours, caption position.
• The deliverables. Every file you expect back.

What do you type into Vizard Agent?

The spec does almost all of the talking here. Vizard Agent needs only two things from your message: the instruction to follow the attached document, and the standing permission to decide for itself — otherwise it stops to ask you things the document has already answered.

Prompt

The spec itself is where the real work is. The strongest rules from a real production document read like this:

• Read the full transcript first. Nothing chosen by sampling.
• Invent nothing. Every number, quote, label or title on screen must have been said in the video.
• Do not stop for approval. Apply the default rule, decide, log it.
• Name the quirks. "The footage records mirrored; do not flip it."

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real episode cut against a channel's own spec file. The document's first rule shapes step eight: read the whole transcript before proposing anything, rather than sampling and guessing.

1. Opens the project and reads the client history.
2. Checks the fonts and resources available on the machine.
3. Downloads the raw video and measures the file.
4. Transcribes the episode and downloads previous deliveries as reference.
5. Reads the complete transcript — the whole thing, not a sample.
6. Detects the face across the video and reviews a contact sheet of frames.
7. Re-measures the face at sixty points with visual validation, then tracks it throughout.
8. Builds a time map accelerating the pauses six times and renders the base video against it.
9. Assembles the audio to the same time map and measures it for normalisation.
10. Selects the overlays against the spec's density and breathing-room rules.
11. Checks every card against the speech that anchors it and adjusts each to its anchor.
12. Lists the profanity with its neighbours and closes the bleep map and the zoom map.
13. Verifies by script that every card's text is literal speech from the transcript.
14. Prepares and renders the animated graphics, composites them over the video and reviews them.

Step thirteen is the one worth stealing outright. The spec forbids inventing text, so Vizard Agent checks programmatically that every word on every card appears verbatim in the transcript rather than trusting itself to have obeyed.

What does the result look like?

The episode this page is written from ran to 170 steps and produced the full set the spec asked for. The spec's rules are visible in the output rather than in the conversation: pauses accelerated on a time map, zooms hitting a stated coverage target, every card anchored to the line that justifies it.

What changes most is the conversation around the work. There is almost none — the document answered the questions in advance, and what comes back is the finished set plus a report of the decisions taken.

When does this not work well?

A spec is only as good as its author, and a bad one is worse than no spec because it is applied consistently. Vizard Agent follows what is written, which means a rule that is wrong gets applied thoroughly.

• Vague rules. "Keep it punchy" is not a rule.
• Contradictory rules. It will pick one and log it.
• A spec that never updates. Revise it when it fails.
• Rules for footage you no longer shoot. Quirks change.
• Nothing about deliverables. Say what files you expect.

How do you fix a result that came back wrong?

Fix the spec rather than the episode. Vizard Agent's report tells you exactly which rule produced the result you dislike, so the correction goes into the document instead of into a message, and every future episode inherits it automatically without you having to remember.

• "The zooms are too frequent." Change the coverage target in the spec.
• "That card is not in the speech." Tighten the no-invention rule.
• "It stopped to ask me." Add the standing permission to decide.

How does Vizard Agent compare to doing it yourself?

By hand, a house style lives in an editor's head and drifts when they are busy or replaced. Writing it down helps a human editor too — but a human still has to remember to check every card against the transcript, and Vizard Agent verifies it by script.

Common questions

What should go in the spec? Tone, hard rules, known footage quirks, the house look and the deliverables. Vizard Agent uses all five.

How long should it be? As long as your rules are. Vizard Agent reads the whole document before starting.

Can I say "don't ask me"? Yes, and you should. Vizard Agent then decides ambiguous cases itself and logs each one.

What if a rule is wrong for one episode? Say so in the spec. Vizard Agent adjusts, delivers something working, and records the deviation.

Does it keep the spec between episodes? Attach it each time to be certain. Vizard Agent treats the supplied file as the authority.

Can I include a "never invent" rule? Yes, and it is the highest-value rule in the document. Vizard Agent verifies card text against the transcript.

What about my footage's quirks? Write them down. Vizard Agent cannot guess that your camera records mirrored.

Can the spec set fonts and colours? Yes. Vizard Agent uses the named typefaces and hex values rather than approximating them.

Should it list the deliverables? Definitely. Vizard Agent produces exactly the files the spec enumerates.

Can I have different specs per format? Yes. Vizard Agent works from whichever document you attach.

Is this just a longer prompt? Functionally yes, but the difference in practice is enormous. A prompt is rewritten weekly and drifts; a spec is a file you revise deliberately, and Vizard Agent applies the same one every time.

How do I start writing one? From your last three corrections. Vizard Agent's report on any episode is a good source of rules you did not know you had.

Will it tell me if it broke a rule? Yes, that is the point of the report. Vizard Agent records every deviation rather than changing your standard silently.