Short version: editing a narrated video is 90% pacing. Cut to a new image every 4–6 seconds on average — faster for action and reveals, slower for a quote or a moment that deserves to breathe — never hold one frame past 12 seconds, and cut the hook tighter than anything else. Keep on-screen text to a few words and only when it adds something. Mix narration on top, music 18–24 dB under it, and a handful of sound effects placed on the exact word. Then watch the whole thing once at normal speed with the checklist below before you upload.
On a talking-head channel, editing means cutting out the ums and tightening the takes. On a faceless channel there are no takes. The narration is a finished performance; the pictures are made to fit it. So "editing" collapses into one question asked a few hundred times: when does the picture change, and to what?
Get that question right and a video made from a script and illustrations feels like television. Get it wrong and it feels like a slideshow with a voiceover — which is what most faceless videos are, and why viewers leave at minute two. Everything below is about that one question.
The baseline for narrated documentary is a new image roughly every 4–6 seconds. That's the rhythm viewers have been trained on by thirty years of television documentaries, and it's fast enough to feel alive without feeling frantic. In practice:
The easiest way to get this right is to cut on the narration's own structure. Every sentence is a candidate for a new picture; every clause with a new noun in it is a candidate too. A good rule for a first pass: one image per sentence, then merge the ones that are visually the same, then add a second image inside any sentence longer than eight seconds.
The editor times every scene to the narration automatically, and the pacing slider — Slow, Medium, Fast, Very Fast — sets how often the picture changes before you draw anything; you see the scene count update as you move it. If you change a line of narration later, the cut re-times itself.
The first 30 seconds are edited differently from everything else, because they are watched differently: by someone deciding whether to stay. Three rules:
Then look at your retention graph after the video has been up a week. If there's a cliff in the first 30 seconds, the hook cut (or the hook script) is the problem — nothing else in the video matters until that's fixed.
A 10–20 minute video needs a second rhythm above the scene rhythm: chapters. Every 2–4 minutes the video should shift — a new question, a new place, a new year — and the edit should mark it: a slightly longer hold on an establishing image, a change in the music bed, a beat of near-silence. Viewers can feel structure even when they can't name it, and a chapter beat is a small promise that something new is coming, which is what keeps them past the mid-video slump.
Add the chapters to the description with timestamps as well. YouTube shows them as key moments in search and viewers use them to skip — which sounds bad until you realise a viewer who skips to chapter three stayed instead of leaving.
Everything above says "cut more". Here's when to do the opposite:
The test for a hold is simple: is the viewer looking at something, or waiting? If they're reading, studying, absorbing — hold. If they've taken it in and the voice is still going — cut.
Text is the most overused tool on faceless channels and the clearest tell of a cheap one. Rules:
The Scene Editor has a lettering fix for exactly the baked-in-text problem: pick the sign or headline, type what it should say, and the crew re-letters it cleanly.
Three layers, mixed in this order:
Use royalty-free libraries — YouTube's own Audio Library is free and safe — and check the licence allows monetised use if you ever plan to join the Partner Program.
The sound designer suggests effects per scene and anchors each to the word it belongs to; levels are set so the narration stays on top. Narration pauses are a marker in the script, so a dramatic beat is one character to add.
Every long video contains two or three Shorts. Finding them is an editing job:
Shorts don't replace the long video — they feed it. A channel that posts one long video and two Shorts a week gets found in two places.
Watch the whole video once at normal speed before you upload — not skimming, not at 1.5× — with this list open:
Expect to find three things. Fix them. That pass is the cheapest quality improvement available to you.
If you're making the video with an AI crew rather than a timeline editor, the division of labour looks like this:
| Editing job | Automated | Your call |
|---|---|---|
| Timing each scene to the narration | ✓ — re-timed when a line changes | Pacing setting (the slider) |
| Choosing the picture for each line | Three directions drawn per scene | ✓ — you pick, swap or redraw every one |
| Hook cut | Draft | ✓ — reorder, tighten, choose the opening image |
| Recurring people and places looking the same | ✓ — cast and places drawn once, reused | Approve each portrait |
| Text in images | Flagged | ✓ — fix the lettering or crop |
| Sound effects | Suggested and placed on the word | ✓ — keep, move, remove, add |
| Music levels | ✓ | Track choice |
| Holds and moments | — | ✓ — entirely a human judgement |
| The final watch-through | — | ✓ — nothing replaces it |
The pattern: the mechanical parts (timing, consistency, levels) are automated; every creative choice is yours and nothing renders that you didn't approve. That split is also what keeps an AI-assisted channel on the right side of YouTube's rules — the channels removed in 2026 were the ones where nobody made any choices at all.
Quick Review arrows through every scene for one-key approve or flag; the Scene Editor swaps, redraws, modifies and re-letters a single scene; the Playback Editor plays the narration and lets you pick between directions live — the closest thing to a real edit bay for a narrated video.
Related: How to start a faceless channel (complete guide) · Best faceless niches · The best faceless video tools.
Every 4–6 seconds on average for narrated documentary: 2–3 seconds for action and reveals, 8–12 for a quote or a moment, and never past 12 seconds. The hook (first 30 seconds) should run faster, at 2–4 seconds per scene.
If you're assembling clips yourself, any timeline editor works — DaVinci Resolve is free and more than enough. If you're using an AI crew like Faceless Channel, there's no timeline: scenes are timed to the narration automatically and you review and choose each picture instead.
Roughly 18–24 dB below the narration. The practical test: if you can hum the track after watching, it was too loud. Duck it under speech and lift it slightly in the gaps and at chapter changes.
On Shorts, yes — they're mostly watched muted. On long-form, no: word-for-word captions pull attention from the pictures. Use short on-screen text for names, dates and numbers only.
By hand, 6–12 hours of the 15–25 a video takes in total. With an AI crew, the timing and consistency are automatic and the human part — reviewing roughly 100–150 scenes and choosing between directions — is about an hour, plus the final watch-through.
Vary the rhythm (fast on reveals, slow on moments), vary framing between consecutive scenes, keep recurring people consistent, fix any text in images, add a handful of sound effects on their words, and watch the whole thing once before uploading. Most "AI look" is actually "nobody made a decision" look.
16:9 at 1080p minimum for long-form (4K if the visuals are sharp enough to deserve it); 9:16 for Shorts, re-framed rather than cropped.
Scenes timed to the narration automatically, three directions per scene, and you approve every shot. Free Starter plan, no card.
Start free — no card →