Endless Generation
It’s never been easier to make another version.
Everywhere I look, another AI tool is promising more.
More images. More video. More voices. More versions. Sometimes unlimited.
And honestly, it’s exciting.
An idea that might have taken days to test can now be explored in minutes. You can try the strange version. Change the camera. Change the performance. Make another frame. See what happens.
That freedom is real. So is the other side of it.
When making one more version costs almost nothing, it becomes very easy to keep making one more version.
Filmmaking used to come with natural stopping points. Film stock ran out. Daylight disappeared. Render times were long. Budgets had limits. Those constraints could be frustrating, but they also forced decisions.
AI removes a lot of that friction. That’s a good thing.
But it also makes it possible to explore long after exploration has stopped helping.
Twenty frames. Nine motion tests. Four performances. Three models. Somewhere in all of that is the version everyone liked — if anyone remembers which one it was.
The problem is no longer just making the thing.
It’s deciding what matters.
Which version was approved? Why did we choose it? What changed? What needs to stay consistent in the next shot?
Generation can be nearly endless.
Human attention is not.
That’s why I keep coming back to production memory.
The goal is not to make AI harder to use. I want these tools to become faster, cheaper, and more accessible.
I just don’t want the ease of making another version to replace the act of choosing one.
Boringly simple.
Extremely serious about human authorship.
We should be free to explore.
We should also know when to say:
That one.
Keep it.
Move on.