ON-CAMERA CRAFT BLOG / 9 MIN READ

How do I improve my talking-head videos?

It is just you and a lens, which means every weak spot is on full display: the light, the eyeline, the voice, the ums. Here is how to fix each one, in the order that actually moves the needle.

3things to fix first
−14 LUFSvoice loudness target
3sto earn the view
0–100craft score

By Thomas, founder of CutScore · Updated June 2026

TALKING-HEAD CHECK · piece_to_camera.mp4
A presenter speaking to camera with a microphone, the classic talking-head setup where light, eyeline and voice decide whether the video holds attention.
CRAFT SCORE
FIXES ADVISED
one person, one lens, nowhere to hide
Eyeline on lens · framing good
Voice a touch quiet · −18 LUFS, lift +400:00
Filler words high · 11 "um" per min02:14
The 30-second answer To improve a talking-head video, fix it in three layers. First the face: put a soft key light off to one side so you are not flat or shadowed, frame yourself with your eyes about a third of the way down, and look straight at the lens instead of your own preview. Second the sound: set your voice near −14 LUFS, keep peaks under −1 dBTP, and kill the background hum, because bad audio sinks a talking-head faster than anything. Third the delivery: open with one strong sentence, cut the dead air between thoughts, and trim the ums. Picture, sound, talking, in that order. If checking all that by hand sounds tedious, that is exactly what CutScore does in one pass.
WHY THIS FORMAT IS BRUTAL

A talking-head video is the most honest format there is. There is no fast-cut montage to hide behind, no sweeping drone shot, no soundtrack doing the emotional work for you. It is your face, your voice, and whatever you are saying. When one of those three is off, the viewer feels it in about two seconds, even if they could not name what is wrong.

I have shipped my share of bad ones. Early CutScore explainer videos where I was lit by a single overhead bulb, talking into a laptop mic, glancing at my own face in the corner of the screen the whole time. I thought I sounded fine. Then I watched it back on my phone on a train, and I looked like a hostage reading a ransom note off a wall just to the left of the camera. The content was okay. The craft was doing everything it could to undermine it.

Here is the good news. Almost every talking-head problem is cheap to fix and has nothing to do with your camera. It is where you put the light, where you point your eyes, how loud your voice sits, and how ruthlessly you cut the dead air. None of that needs new gear. Below is the order I would fix them in, starting with the one viewers notice first.

THE FIX LIST

The seven things that make or break a talking-head video.

Each one has a target you can actually hit, and each one is something a viewer will clock if you get it wrong. Fix them top to bottom.

What to fixTarget to hitWhat it costs you if you skip it
Eyelineon the lensLooking at your own preview reads as shifty, like you are talking past the viewer.
Light on your facesoft, from one sideFlat overhead light makes you look tired; one hard shadow makes you look sinister.
Framingeyes on upper thirdToo much headroom or a chin-up phone angle looks accidental and amateur.
Voice loudness≈ −14 LUFSA quiet voice feels timid next to every other video in the feed.
Background noiselow, steadyRoom hum and reverb read as "cheap" before you finish your first sentence.
Filler wordsa few, not a floodA dozen ums a minute quietly tells people you are not sure of yourself.
Pace and the hookcut dead air, open strongA slow opener and long pauses send people straight back to the scroll.
The one most people get backwardsYou do not need a better camera. A clean phone shot with a soft light, a close mic and a tight edit beats a cinema camera in a dark room with laptop audio every single time. Spend your effort on light, sound and delivery, not megapixels.
SKIP THE MANUAL CHECK

Checking your light, eyeline, loudness and filler-word count by hand on every take adds up. CutScore measures all of it in one pass and hands back the fixes, so you can record again instead of inspecting.

Join the waitlist
HOW TO ACTUALLY FIX EACH ONE

Three passes: face, voice, talking.

1. The face: light, eyeline and framing

Start with the light, because it does the most for the least money. One soft source, off to one side and slightly above your eyes, beats any expensive lens. A window with a sheer curtain works. So does a cheap softbox. The goal is gentle shadow that gives your face some shape, not the flat under-the-bulb look that ages everyone by ten years. Then the eyeline. Look at the lens, not at the little preview of your own face, because viewers read eye contact instantly and a sideways glance feels evasive. I stick a small arrow next to the lens so my eyes have a target. For framing, put your eyes roughly a third of the way down the frame, leave a little headroom but not a swimming pool of it, and get the camera up to eye level so you are not shooting up your own nose.

A hand riding the faders on an audio console, a reminder that the voice in a talking-head video matters more than the camera and needs to sit near -14 LUFS.
In a talking-head video the voice carries the whole thing, so it gets the same care as the picture. Photo: Tima Miroshnichenko / Pexels.

2. The voice: the part that actually carries the video

In a talking-head video the voice is the video. People will forgive a slightly soft image. They will not forgive a voice that is quiet, boomy, or buried in room hum. Get the mic close, within a foot or two of your mouth, because distance is what lets the room creep in. Then handle two numbers. Loudness, which you want sitting near −14 LUFS for YouTube so your voice does not feel timid next to the next creator, and true peak, which you keep at or below −1 dBTP so nothing crackles after the platform re-encodes your file. If you can hear hiss or a fridge hum in the background, a light noise-reduction pass clears most of it. The test is simple: play it back on your phone speaker at half volume. If every word is clear, you are done.

3. The talking: filler words, energy and the hook

Now the delivery, which is where most talking-head videos quietly leak attention. Count your filler words: a few ums are human, but a dozen a minute tells people you are improvising and not sure where the sentence is going. Record a couple of takes, talk with slightly more energy than feels natural (the camera flattens you, every time), and keep the take where you sound like you are talking to a friend. Then think about the opening. Watch your first three seconds as if you were thumbing past it in a feed. If you open with "hey guys, so today I wanted to talk about," you have already lost half of them. Lead with the most interesting sentence you have. The introduction can come after you have earned the view.

4. The cut: give a static frame some life

One person, one held shot, no movement: that is the recipe for "boring," even when the content is good. The fix is not a film crew. It is the edit. Cut the dead air between your thoughts with a jump cut so the pace stays tight and you never sit there watching yourself breathe. Drop in the occasional cutaway or a relevant graphic to give the eye something else to land on. Keep the average shot length shorter than feels comfortable, because what feels fast to you after forty viewings is normal to someone seeing it once. The honest test: would you keep watching this if it were not you on screen?

RATHER SEE IT THAN READ IT?

Here is a real CutScore coaching report: eyeline, light, loudness, filler words and pace, all scored, with timestamps and the exact fixes.

See a sample report
SHORT ON TIME

If you only fix three things.

Most of the jump from "homemade" to "this person knows what they are doing" comes from these three. Do them before your next recording.

1
2-MIN FIXLIGHT
Put one soft light to the side of your face
Move your main light off-axis and slightly above your eyes, soften it with a curtain or a diffuser, and turn off the flat overhead bulb. That single change gives your face shape and instantly reads as "lit on purpose" instead of "filmed wherever I was sitting."
How A window with a sheer curtain on one side works for free. Sit so the light falls across your face, not flat into it.
2
AUDIOSOUND
Set your voice to about −14 LUFS
Get the mic close, then normalise the whole mix toward −14 LUFS with a true peak under −1 dBTP. A quiet talking-head feels weak the moment it sits next to anything else in the feed, and it has nothing to do with how expensive your mic is.
How Run a loudness meter over your export, or let CutScore measure it and tell you the exact gain change.
3
EDITDELIVERY
Look at the lens and open strong
Stop watching your own face in the corner of the screen. Look straight at the lens, and cut the slow introduction so your first sentence is the most interesting thing you have to say. Eye contact plus a strong opener does more for retention than any new piece of gear.
How Stick an arrow next to the lens as a target. Re-cut the opening so the payoff lands before second three. See the hook.
THREE WAYS TO CHECK YOUR OWN

By eye, by meter, or in one pass.

OPTION 01

Watch it back honestly

Free, and better than nothing. Watch on your phone, at normal brightness, a day after you record so you are not still in love with the take. Check the list above against targets, not vibes. The catch: after forty viewings, your own ear stops hearing the ums.

OPTION 02

With meters and a second opinion

Accurate, but slow. A loudness meter for the voice, a scope for exposure, and ideally a friend who will tell you the truth about your eyeline and your pacing. Great if you have the tools and the patience. Most creators recording weekly do not.

OPTION 03

With a coach in one pass

Hand the file (or a link) to CutScore. It measures light, framing, loudness, filler words, pace and the hook against the right standard for a talking-head, then gives you a 0 to 100 score with timestamped evidence and the fixes. See a sample report.

How CutScore scores a talking-head video CutScore is an AI video quality coach for pre-publish QC. It computes the measurable craft deterministically (loudness with an EBU R128 meter, exposure, framing, shot length, filler-word density) and reserves AI for the genuinely subjective parts, like whether your opening actually earns the view. You get one score, the evidence behind it, and a prioritised list of fixes, before anyone else watches. It judges the craft of the video itself, not your tags or thumbnail, so it sits next to a growth tool rather than competing with one. More on the method and the standards.
QUESTIONS

Frequently asked.

Fix the face first: get a soft key light off to one side, frame yourself with your eyes a third of the way down, and look at the lens, not your own preview. Then fix the sound: set your voice near −14 LUFS, keep peaks under −1 dBTP, and cut the background hum. Finally, tighten the talking: open with one strong sentence, cut the dead air between thoughts, and trim the ums. Picture, sound, delivery, in that order.
Usually it is one held shot, flat light, and a slow opening. A static frame with no movement and no cuts gives the eye nothing to track, so attention drifts. Add a little energy in the delivery, cut the pauses with jump cuts, drop in one or two cutaways, and open with the most interesting thing you have to say instead of a slow introduction.
Look directly at the lens, not at the screen showing your own face. Viewers read eye contact instantly, and looking at your preview makes you appear to glance just past them, which feels evasive. Put a small arrow or sticker next to the lens as a target, and if you use a teleprompter, keep the text as close to the lens as you can.
Talk slightly faster and with a touch more energy than feels natural, because the camera flattens you. Record a few takes, keep the one where you sound like you are talking to a friend, and cut the rest. Reducing filler words helps too, but do not over-edit the life out of it. A little imperfection reads as human.
EARLY ACCESS

Stop guessing what is holding your face-to-camera back.

CutScore checks your light, eyeline, voice, filler words and pace, then tells you exactly what to fix, with the evidence to back it up. Join the waitlist for early access.

Join the waitlist