How to Lay Soft B-Roll Under a Screen-Recording Narration

Anyone editing a product lesson can lay soft b-roll under a screen-recording narration in 12 minutes: keep the cursor readable, duck the bed under speech, and never let stock footage steal the click. Do it in CapCut, Descript, or Runway before a "cinematic underlay" turns a tutorial into a trailer.

Soft b-roll is a bed, not a second hero. If the viewer cannot see which button you pressed, you never taught. You decorated.

This is not b-roll that matches a talking-head. That skill pairs cutaways to a face. This skill pairs a quiet bed under a UI. It is not cutting dead air from a talking-head without fake jumps. Dead air is a pause in a face. Here the voice sits on a screen. It is not freezing a talking-head frame as a title card. A card is a still before the lesson. Soft b-roll lives under the lesson. Save the duck-plus-opacity as a skill prompt you can rerun.

Lay soft b-roll so the screen still teaches

1. Mark the screen as primary. Name the bed

Play the screen recording once with a pad and a pen. Mark the moments the cursor must stay obvious: first click, first dropdown, first error, first success.

Write three facts on paper: how long the take is, how many soft underlay clips you will allow (a number, not "make it pretty"), and whether the VO already names what the cursor is doing.

If you cannot name those three, watch again. A seven-minute ramble is still one lesson.

Banned: cinematic underlay, epic stock, full-screen takeover, whip pan over UI, beat-sync the cursor, neon overlays, auto-duck as a personality, trailer energy, hide the boring screen, make it pop.

2. Choose soft clips. Reject loud ones

Soft means low contrast, slow motion or locked-off, no big faces stealing focus, opacity often between 15% and 35% under the screen. Office blur, soft gradient, shallow desk detail, gentle hand on notebook — only if the UI stays king.

Reject: city drones, fireworks, hard zooms, text-heavy stock, another product UI that competes, anything with a bright logo in the corner.

Instruct yourself: "Bed supports the VO. Screen carries the proof. If I mute the VO and the bed still feels like the A-roll, the clip is too loud."

Good bed: soft desk bokeh at 25% under a settings panel. Bad bed: a full-screen coffee pour that covers the Save button for two seconds.

3. Duck under speech. Leave air in the pauses

Lay the bed on a track under the screen. CapCut, Descript, or Runway — same rule: when the mouth (or VO) is talking about a click, the bed ducks. When there is a half-second breath, the bed can rise a little. It never outruns the voice.

Hard numbers you can steal: bed peaks around −24 to −30 LUFS under clear VO, and never masks consonant clicks. If your tool only has a duck slider, duck until you can still hear "click Save" without rewinding.

If the VO is thin, fix the VO. Do not raise the bed to "fill energy." Energy is how tutorials become ads.

4. Protect the cursor. Force a mute-on pass

Mute the VO. Watch the timeline once. Can you still see every critical cursor move? If a soft clip covers a dropdown at 40% opacity, lower it or cut a hole (mask / picture-in-picture crop) so the control stays clear.

Tell the editor: "Screen stays above 70% visual priority on every instructional frame. Soft b-roll never owns the click."

ChatGPT will not save you here if you are prompting AI b-roll: if you generate underlay in Runway, prompt "soft out-of-focus desk texture, no faces, no text, slow, loopable" and still check the cursor. A generated neon desk is still loud.

If a clip has a hard cut that jumps under a sentence, dissolve 8–12 frames or cut on a breath. Soft includes the join.

5. One human pass as the learner who never saw the product

Watch once at 1x with sound, as if you must click the same path tomorrow without notes.

Check three traps:

• A bed that teaches a different mood than the VO (trailer music under a tax setting).
• A cursor lost under opacity.
• A pause where loud b-roll suddenly becomes the show.

Fix those. Then stop. Export. If you still want a second neon layer, you are making a trailer. Stop.

Reusable skill card

Trigger: You are about to drop stock footage under a screen recording and type "make it more dynamic."

Input: Screen take, VO, max clip count, opacity range, duck target, plus the banned-phrase list.

Output: Soft underlay that ducks under speech. Cursor readable. No full-screen takeover.

Stop: After one mute-on pass and one learner pass. No louder bed. No whip over UI. No second hero.

Takeaways

• Soft b-roll is a bed. The screen is the lesson.
• Duck under speech. Protect the cursor. Reject loud stock.
• Mute once to verify every critical click stays visible.
• If the VO is weak, rewrite the VO. Do not bury it.