Video Podcast Eye Contact: What to Fix When You Publish Every Week
Video podcasting put a camera on a format that was designed for none. Radio habits — notes spread out, eyes on the question doc, full attention on the guest — are all still correct for the conversation and all wrong for the lens. The result is a host who spends most of an hour looking somewhere other than the audience.
The honest answer is that this matters much less than tool marketing suggests, and in some segments it does not matter at all. This is a guide to telling those apart, because correcting an entire ninety-minute episode is both expensive and unnecessary.
Where gaze actually costs you
In conversation, looking at your guest is correct. Viewers understand the grammar of a two-person interview instinctively and do not expect either party to stare down the barrel while the other is speaking. Correcting that footage would make it stranger, not better.
The segments that carry a real cost are the ones where you are addressing the audience directly — and those are also the segments doing the most commercial work on your show.
- The cold open and the first thirty seconds, which decide whether anyone stays.
- The intro read, where you explain what this episode is and why it is worth an hour.
- Sponsor reads, which is the one place advertisers explicitly care how present you look.
- Solo episodes and monologue segments, which are entirely direct address.
- The outro and call to action, where you are asking for something.
- Clips cut for social, which are almost always taken from the direct-address moments.
Correct the segments, not the episode
Adding up the list above gives you perhaps six to ten minutes of an hour-long show. That is the number to work with, and it changes the economics completely.
At roughly $0.35 per minute, correcting eight minutes of direct address is under three dollars an episode — around $150 across a year of weekly publishing. Correcting all fifty-two full episodes would be an entirely different and largely pointless expense, because most of that footage is conversation that should stay as it is.
The workflow is to cut your episode first, export just the direct-address segments as their own files, correct those, and drop them back into the timeline. It is one extra step in an edit you are already doing, and it targets the money precisely at the parts of the show that need it.
The setup changes that pay off across a whole season
Because a podcast is a repeating format, anything you fix in the setup gets fixed for every future episode. This is the highest-leverage half hour you can spend on the problem.
- Put your question doc directly under the camera, not on a second monitor beside it. Most hosts have this the wrong way round because the second monitor is bigger.
- Raise the camera to eye level. Podcast desks tend to put it low, which produces the least flattering and least correctable angle.
- For remote guests, place their video window immediately below your webcam, so looking at them looks almost like looking at the audience.
- Cut the question doc down to keywords. A full script guarantees your eyes stay down; five words per question lets you look up.
- Mark your sponsor read in the doc as a note to look at the lens. Knowing it is coming is usually enough.
Where correction will not help
- Wide two-shots where both people are small in frame. There is not enough eye region for the model to work with.
- Footage where you are turned toward a guest sitting at ninety degrees. That is a head turn, not a gaze offset.
- Long stretches of reading. Correction is convincing in short bursts and starts to look uncanny across ten unbroken seconds of looking away.
- Heavy glasses glare, which sits exactly where the pupils need to be visible.
- The conversation itself — which, again, is usually better left alone.
Why per-minute pricing suits a podcast
Podcast production is relentless but small in per-episode volume once you are only correcting direct address. That is an awkward shape for subscriptions, which assume either heavy monthly use or none, and a natural shape for credits.
GazeFix charges per minute of video with no subscription, credits never expire, and exports are watermark-free at original resolution — which matters when the corrected segment has to intercut invisibly with uncorrected conversation footage shot on the same camera. A video that fails to process is refunded automatically.
Signing up gives you one free credit, no card required. The useful test is your last sponsor read: correct that single minute, drop it back into the episode next to the untouched footage around it, and see whether the join is invisible. That tells you more than any demo reel.
Vertical clips change the calculation
Most video podcasts now get more reach from thirty-second vertical cuts than from the full episode, and gaze behaves differently once the frame is cropped. A wide two-shot that read perfectly well at 16:9 becomes a tight head-and-shoulders crop at 9:16, and a face filling most of a vertical frame makes off-camera eyes far more obvious than the same footage did in the wide.
This cuts both ways. The crop is why gaze suddenly matters on clips that looked fine in the episode — but it is also why correction works better there, because a large, well-lit face is exactly the input the model handles best.
The practical approach is to correct at the clip stage rather than the episode stage. Cut your vertical first, then correct the cut, so you are paying for thirty seconds rather than ninety minutes and correcting a crop where the eyes are large enough for the result to hold up.
- Cut the vertical clip first, then correct it — not the other way round.
- Prioritise clips where you are speaking to camera over clips of your guest answering you.
- Check the first second specifically. Vertical feeds are judged in a scroll, and that frame does the work.
- Watch for correction on a face that has been scaled up heavily from a wide shot; upscaled eyes correct less convincingly.