The Loud Part of a Quiet Job

Everything else we make is an argument for quiet. The thumbnail and the title are the one part of the job where we have to behave like everybody else, and they are the first thing anyone ever sees of us.

The conflict is real, not a matter of taste

It would be comfortable to say this is a false dilemma and that good work finds its audience. It is not, and it doesn't.

A thumbnail is competing in a grid, at about the size of a postage stamp, against video about arguments, disasters, food and money. A person scrolling is making a decision in well under a second and is not reading anything. In that environment, contrast wins, faces win, text wins, and a dim brown rectangle of a room at night loses to all of them without a fight.

So the honest statement of the problem is: the thing that makes an episode good to watch is exactly the thing that makes its thumbnail invisible. Low contrast, no faces, no focal point, nothing at stake. We are not being punished for a lack of craft. We are being punished for the product.

Every choice that makes the episode calmer makes the thumbnail quieter, and the thumbnail is the only part of it most people will ever see.

Three generations, and what the numbers did

We have rebuilt our thumbnails three times. Here is what happened, including the part where we were wrong.

1. 1

   A frame from the episode

   The purist approach: pull a still, crop it, done. Completely honest — what you see is literally what you get. It also performed terribly. Around three in a hundred people who saw it clicked, and the ones who did were almost all people who already knew us. As a way of reaching anyone new it did not function.

2. 2

   A frame, pushed

   Same image, but treated: a warmer grade, the lamp brightened, the darks lifted so it survives being a postage stamp on a phone in daylight. About five in a hundred, so a little over half again as many people. This is where most of the improvement came from, and it is worth saying plainly that it came from legibility, not from drama.

3. 3

   Painted, with the duration on it

   Where we are now. A composed illustration rather than a frame, built for the small size, with the running time set in a corner in the same typeface the site uses. A bit under six in a hundred. The gain over the second generation is small and the effort is much larger, but the duration label turned out to matter more than the picture — people want to know what they are agreeing to.

The thing we got wrong in between: for a few months we tested brighter, higher-contrast versions with the colour pushed well past anything in the episode. They performed best of all — around seven in a hundred — and the watch time behind them was noticeably worse. People arrived expecting one thing and got another, and left in the first minute. We stopped, and the number we lost by stopping is roughly one person in a hundred, permanently.

The things we don't use, and what they cost us

There is a standard toolkit. It works. We have tested most of it.

The cost is not hypothetical. Our click-through rate is roughly half what a competently aggressive channel in this space gets, and about a third of what the most aggressive ones get. That is not a moral victory, it is a smaller audience — fewer people finding something they would have liked. We have made peace with it, but we do not pretend it is free, and anyone who tells you restraint costs nothing is selling something.

Where we have overstated

The promise of this piece was honesty, so: some of our thumbnails are more dramatic than the episode behind them, and that is a form of exaggeration even when nothing in the image is false.

The first eight-hour rain window. The thumbnail has a lightning flash in the far distance. There is exactly one comparable moment in eight hours, in the third hour, and it is much dimmer. We drew it because the image needed something to look at. We would not draw it now.

The winter-hearth long version. The fire in the thumbnail is roughly twice the size of the fire in the episode. Nobody has ever complained. It still bothers us, because the whole point of that episode is a small fire.

The first harvest-day episode. The sky is a sunset. The episode is overcast for most of its length and the sun only breaks through near the end. This is the one that most clearly crosses from treatment into misrepresentation, and it is the reason we now build the thumbnail from a shot that actually occurs.

The rule we settled on afterwards is narrow enough to be checkable: the thumbnail must depict a moment that is in the episode, at roughly the intensity it occurs. Warm it, light it, compose it, paint it — but do not invent the weather. We break this less than we used to. We have not broken it since the harvest episode.

Titles: eight hours means eight hours

Titles are the easier half, because there is a bright line available and we can just stand on it.

If the title says eight hours, the file is eight hours. If it says rain, it rains for the duration and does not fade into wind for the last two. If it says a fireside, there is a fire visible, not implied by light on a wall. The description says what is actually in it, including the parts that might disappoint you — whether there is music, whether there is a cat, whether the rain has thunder in it.

This sounds like a low bar. In this genre it is not. A large share of long ambient video is padded, cross-faded from a different scene, or half the length the title claims once you subtract a silent tail. Duration is the main promise the title makes — this will not run out before you do — and it is the easiest one to quietly break.

We put the running time on the image for the same reason. It is the single piece of information a person actually needs at that moment, and putting it there costs us nothing but the drama of not saying.

Where we have landed

Not a resolution — a permanent arrangement between two parties who want different things.

The platform wants a picture that interrupts you. The room wants a picture that doesn't. We give the platform legibility, warmth and a clear duration, and we withhold the arrows, the capitals and the invented weather. Somewhere in that split is a number of people who never find us, and we cannot see them, and they are real.

What we can do is make sure that the people who do arrive get what the picture said. Our covers live on the episode shelf, and how the episodes behind them get built is in how an episode is made.

The evening is warm here.