The Two-Column Script Format for YouTube: An A/V Script Template for Faceless Videos

A two-column script for YouTube is a table: the left column carries the narration word for word, and the right column carries everything that is not spoken, from visuals and on-screen text to sources and sound cues. Faceless, documentary and video-essay channels use the format because it hands an editor, a B-roll shooter or an image workflow an ordered shot list without a meeting.

What is a two-column (A/V) script?

A two-column script, also called an A/V script, is a script laid out as a table with one column for audio and one for video. Each row is one moment in the finished video. Read down the left column and you have the narration; read down the right and you have the shot list.

The audio column holds the narration exactly as it will be spoken, because the narrator or the voice model reads the cell, not the intention behind it. The video column holds what happens on screen while that line plays: the footage or image, the framing, and where the asset comes from. The minimal shape:

It is also called the audio-visual (A/V) script format, and faceless channels with no host use it as their default. Everything past two columns refines the visual side: on-screen text, sources and sound get their own cells, so editor, thumbnail designer and fact-check read one table.

When should you use one instead of a plain script?

Use a two-column script whenever someone or something other than you has to turn the words into pictures, and stay with a plain script when you record and cut your own talking-head footage in one pass.

The format earns its keep in these cases:

• •A freelance editor assembles the video from your narration. The visual column replaces the "what should I show here?" message thread.
• •The video depends on B-roll you have not shot yet. The visual column is the shot list, already ordered.
• •The visuals come from an AI image workflow. Each row becomes one generation prompt with its on-screen text already decided.
• •On-screen text carries information the narration does not say out loud. A dedicated column keeps it from being improvised during the edit.
• •More than one person touches the script: writer, narrator and editor work from the same rows instead of three documents.

A plain script wins when you are the host, the footage is you at a desk, and the edit is cuts between sentences; extra columns are ceremony there. For the two most common faceless formats, start from the faceless narration template or the video essay template and keep only the columns your editor actually reads.

How do you fill the visual column?

Fill the visual column in the order the narration runs, one row per beat of narration, with a description an editor can shoot, buy or generate without asking you what you meant.

Five rules that hold across workflows:

1. •Break rows at beats, not sentences. A new idea, example or location starts a new row; two sentences that share one picture share one row.
2. •Name subject, framing and movement. "Scale inside the water tank, macro shot, slow push-in" can be produced. "Show the machine" cannot.
3. •Tag where each asset comes from. HAVE for footage in hand, SHOOT for the shot list, STOCK for the search that still has to happen, GENERATE for images a model will make. Editors triage the video by origin.
4. •Write fragments, not prose. The cell is a brief for someone else, not a paragraph for you.
5. •Keep one style note and reuse it. For generated images, a fixed style line in front of every prompt is what stops thirty images from looking like thirty different videos.

One tool fills this column from the narration: Prepublish's script-to-image-prompts tool takes a pasted script and returns one plain-English image prompt per scene, in script order, each quoting the lines it covers. The page documents one prompt per 60 to 120 words of script, with every prompt naming subject, camera angle, lighting and style. It also returns three thumbnail concepts, each with the text overlay it was written to carry. The prompts are model-agnostic, so the same list works in ChatGPT, Midjourney, Flux or any other image model, and the tool writes prompts only: it generates no images.

How do you mark sources and on-screen text?

Give on-screen text and sources their own cells, so an editor can tell a graphic that must be built from a citation that never appears on screen.

• •On-screen text goes in quotes, exactly as it should appear. "DESCALING, STEP 2 OF 4", not "something about step two".
• •If the text carries a number, its source sits in the same row. A statistic on screen with no row-level source is a correction waiting to happen.
• •Record sources well enough to find them again. Page title, publication, and the link pasted once. The cell serves whoever builds the citation card or answers the comment asking where a figure came from.
• •Keep sources in the table even when nothing appears on screen. The column is the paper trail for the claims the narration makes, and it shows which rows need a source card after all.

One row, one claim. If the narration in a row makes two claims, split the row, so each source stays attached to the sentence it supports.

A copy-paste two-column template

Copy the table below and replace the cells. It is still a two-column script: everything right of the narration column is the visual side, split out so nothing is lost in a handoff. The rows show one short segment from the opening of a faceless video about script formats.

To turn this into a blank template, keep the header row, delete the example cells, and add one row per narration beat, leaving Sound empty wherever the row needs no effect. Google Docs Help documents the steps for building the table as Insert, then Table, then choosing how many rows and columns to add, with tables as large as 20 x 20 cells (Google Docs Help).

You do not need all six columns for every video: a short one runs on Audio, Visual and On-screen text, with sources collected in the description document until an editor or a fact-check joins the workflow.

How do you check the narration column before recording?

Check the audio column on its own, before any visuals are attached: once the video column is full, a thin script looks finished.

• •Read the audio column aloud and time it. The cells are what a narrator or voice model will say; the runtime comes from that word count, not the visual column's ambition. The script length to runtime calculator turns that word count into a runtime at the median pace for your YouTube category, and back again when you start from a target length.
• •Judge the opening as text. The first rows must state the premise and give a reason to keep watching; no picture rescues an opening that never says what the video is. Work through the script structure guide before freezing the table.
• •Watch for cells doing the narration's job. If a row's meaning lives only in the visual column, the audio alone will confuse; rewrite the row so it survives on sound.
• •If a synthetic voice reads it, write for the voice. A voice model reads the cell exactly as written, so sentence length, numbers and names need the AI voiceover rules applied before the render.

Then check the script itself. Prepublish (prepublish.ai) checks a YouTube script before you record: it marks the passages most likely to lose attention. It is independent of YouTube and is not YouTube Studio's "prepublish checks" feature.

Before you record, paste the narration column into the free script check at prepublish.ai and see which lines carry the most attention risk. The check includes a policy pre-flight against YouTube's published guideline categories, and needs no login.