What I’m missing isn’t getting a clip to come out right. That part works now. What I’m missing is getting one piece to join up with the next: the giraffe in the second piece being the same giraffe, on the same table, under the same light.
And it turns out there’s a shortcut. If both shots come out of the same generation, there’s nothing to join.
What the manual said
Google publishes a guide on how to ask its model for things, and it says you can fit several shots into a single clip by writing the request in time stretches: second zero to three, wide shot; three to six, mid shot; six to eight, the face.
That sounded like solving both my problems at once. I gave myself three attempts and two and a half euros.
The three attempts
The first one, exactly as the guide puts it. The three framings did come out, but chained into one continuous push in.
The second, writing “hard cut” in capitals, which is what half the internet recommends. One single jump, and late at that: where I’d asked for a cut on second three and on six, there was one at five and a half.
For the third I stopped giving instructions and described a shoot to it instead. Three cameras, each in its own place, the first one locked on a tripod, and one sentence at the end: the shots change all at once, like a film that’s been edited.
That did it. Two real cuts, almost where they belonged.
What I take away, which isn’t what I went in for
I went in for the shots and what I found was continuity.
In all three attempts, the bad one included, the giraffe is exactly the same in the three framings. Same eyes, same markings, same table, same light.
Not one complaint from the quality check about that, and it does like to make them.
That’s half of what I’ve been chasing for two weeks, and here it comes free, because there aren’t two generations to square up: there’s one.
What still doesn’t work
The timings are a suggestion. The cuts land half a second before you ask for them, so if you want a cut to fall on a beat of music, you’re not going to nail it in writing.
And as the shot closes in, the real world creeps in: in the mid shot a strip of furniture shows up behind that you couldn’t see in the wide.
The photo the clip starts from is a giraffe on the table at home, and home is right there, waiting for the camera to come closer.
Three attempts, two and a half euros, and one step up.
Until today, to tell a scene in three shots I had to generate three times and pray the toy didn’t change along the way. Now I know how to ask for one.