Fire Is the Hardest Thing to Draw

That is the whole problem, and it took two years and a lot of bad flames to state it plainly.

The familiarity threshold

There is a line running through every element you might animate, and it is drawn by the viewer's experience rather than by physics.

Below it sit the things people have seen but never studied. Rain through glass: everyone has looked at it, almost nobody has looked at it. So you can invent a rain that behaves nothing like real rain — slower, more evenly spaced, drops holding their shape far past what surface tension allows — and it reads as rain. It reads, in fact, as nicer rain.

Above the line sit the things people have watched deliberately, at close range, while doing nothing else. A face. Hands doing work. An animal. And fire — the only one of those watched purely for its own sake. Everyone has at some point sat in front of a fire and watched it do nothing for twenty minutes. That is undirected, extremely close observation, and it builds a model in the head that no viewer can articulate and no animator can fool.

When a flame moves wrongly, nobody can tell you what is wrong. They just say it looks fake, and they are always right.

This is the cruel part. Bad rain gets a note you can act on. Bad fire gets a shrug, and the note arrives as a verdict on the whole scene — the room feels off, the episode is somehow colder. We have spent whole review sessions circling a problem that turned out to be nine frames of flame rising at a constant speed.

Three layers, and only one of them may lie

We build every fire in three layers, because they behave like different materials and the eye judges each by a different standard.

1. 1

   The base — the part that must be true

   Coals, the low sheet of flame sitting on the fuel, the glow inside the gaps. Nearly still, brightest thing in frame, and where the eye actually rests. Every attempt to stylise it has failed. It must sit on the wood, be brighter where the wood has burned through, and change so slowly that a minute of watching tells you only that it has changed, never when.

2. 2

   The tongues — the part that gets you caught

   The visible flames — the most tempting layer to design, because they have shape and shape is fun. Also the layer where the viewer's model is most precise: flames accelerate as they rise, thin, detach, never move at a constant speed and never repeat a shape. Almost every failure of ours lives here.

3. 3

   The sparks — the part allowed to lie

   Embers lifting off the fire, and the one layer where you can invent freely. Real sparks are sparse, dim and irregular; ours are more frequent, warmer and slightly slower, and nobody has ever objected — because nobody has ever tried to track a single spark. It is the only part of a fire the eye samples rather than watches, so the only part you can compose.

The working rule comes out upside down: the calm layer must be accurate, the dramatic one may be fiction.

The real work is on the wall

We understood this far too late: most of the believability of a fire is not in the fire.

A fire is a light source that moves, so everything in the room is lit by something unstable — the opposite wall brightening and dimming, shadows swinging by small amounts, a shelf catching a warm edge that comes and goes. If the flame flickers and the room does not, the brain files the flame as a picture pasted onto a photograph. It does this instantly and without asking.

The correction is unglamorous. We animate the room light first, from the same curve that drives the flame, and draw the flame into it afterwards. When we have to choose — and on a tight week we do — the moving room light wins. A stiff flame in a room that breathes beats a beautiful flame in a frozen one.

The single most useful thing we ever added to a hearth was not a flame at all.

Under everything sits a slow cycle we call the breath: about ten seconds long, applied only to the brightness of the coal bed, rising slightly faster than it falls. It is subtle enough that you cannot catch it happening even when you look. But switch it off and people describe the room as colder — not darker, colder — which is strange, consistent, and something we have never fully explained.

Our best guess is that it does the job room tone does in a sound mix: it tells you the thing is alive and keeps a rhythm of its own, unsynchronised to anything you are doing. A fire exactly as bright now as it was thirty seconds ago is a lamp.

Where ours are still wrong

We have not solved this. The list below is the archive as it stands.

The early fireside episode. The worst one, and it stays up. The flames sit in front of the hearth rather than in it — the layer order was wrong and nobody caught it before the render. Once you see it, the fire is a sticker. It is also on a twelve-second loop with a cross-fade, so the brightness dips at every join.

The first winter-hearth episode. Flames acceptable; room dead. The walls carry a fixed warm gradient painted on rather than light driven by the fire, so a candlelit room sits next to a moving flame with no relationship between them. This one taught us the wall lesson, at the cost of six weeks.

The lantern workshop, one shot. The stove in the background of the wide has a flame with a uniformly bright base — a paper cutout with a lamp behind it. Small, but one of us cannot watch that shot.

We are not going back to fix them. Re-rendering old long files to improve a flame is a great deal of machine time spent on the past, and we would rather the past sit there honestly. The early episodes are worse than the current ones. That is what an archive is.

When we simply don't show it

There is one more option, and it is a cheat, so we should name it as one.

Quite often the fire is off-screen: light on the wall, a log settling, an orange edge on a chair arm, no flame in frame. It works beautifully, and it is unmistakably a way of avoiding the hardest thing in the shot.

We allow it when the fire is not the subject, and refuse it in an episode that is about a hearth — at that point we would be selling something we declined to make. The test: would we still choose it if the flame were easy? Honestly, about half the time, no.

The part we have not worked out

There is no ending here where we hand you the technique that solved it. Fires are still the shots we schedule the most time for and finish least happy with, and the gap between our best flame and a real one is visible to us in every episode.

What changed is where the effort goes. The flame is no longer the thing we fuss over; the room around it is. Told that in the first year, we would not have believed it, and we would have spent another six weeks on tongues.

The rest of how an episode gets built is in how an episode is made.

The evening is warm here.