On 10 August I wanted to stage a shipwreck in the pool at home: a toy ship, a leak and the ship going down, in three shots.
Twenty-five generations later I didn’t have a single one I could accept.
It’s not that they came out so-so. They failed somewhere different every time, and every fix of mine broke something else. The ship wouldn’t sink: it stayed afloat with a dry deck, drifting slightly back in the frame. A dark thread with a little stick hanging off it came down from the keel, and it screamed that the ship was being dragged. The camera wouldn’t hold still either. The wooden door in the background drifted out of shot and rooftops came in that weren’t there at the start.
What I was doing was making it worse
With every fault I wrote more and forbade more things, since it was the only thing I could think of. No camera movement. No sails.
But forbidding a thing puts it in the shot. The third shot came out with sails all three times I expressly forbade them, and the camera did the same: it was banned from moving and it moved anyway. The model doesn’t read a “no” as a wall. It reads the word that comes after it and puts it in the shot.
Making the request longer to fix a fault doesn’t work either. I stopped doing it the day I counted the damage: three clips that were halfway there ended up discarded entirely after I improved them.
Scale was the last warning. Seven takes in a row with the figure the size of a building, while I insisted in writing that it was a small toy and the photo I’d handed over said otherwise.
The only rule I take seriously
That batch produced the one thing that has moved me forward since. It isn’t a technique, it’s a notebook.
Every take gets one line with its verdict and its reason: red, amber or green, what fails and at which second. I have forty takes written down. Thirty-five are red, which isn’t a pleasant number and is exactly why I leave it in writing.
The rule comes with the notebook. A change in how I ask that doesn’t lower the percentage of reds gets reverted, whether I like it or not and however good the explanation sounds. Without that number, every afternoon of testing ends in a new theory about what the model likes, and no way to know if it’s true.
I didn’t invent this here. It’s what we did with a ship’s final quantities: with no numbers written down there’s no argument, there are opinions. The difference is that there a mistake moved tonnes of product in the wrong direction, and here it costs me an afternoon and a few cents.
The video below is that same ship, in that same pool, in a shot that did work. But what fixed it wasn’t better writing, and that goes in the next report.