Choosing Your AI Stack

I spent my twenties convinced the next lens would fix my films. It never did, but the disease was at least visible — a shelf of glass I could look at and feel ashamed of. The AI version is sneakier. It's six subscriptions auto-renewing in the background, each one justified by a project you haven't started, each one costing less than dinner so you never quite cancel. Subscription bloat is gear-lust with better camouflage, and it drains an indie budget just as efficiently.

The cure is the same as it ever was: stop collecting capability and start matching tools to jobs. Not "this could be useful someday" — this does a task on the film I'm making right now. That single shift kills most of the bloat before it starts.

The rules of a lean stack

• One tool per job. One writing assistant, one image generator, one video tool if you even need one yet. Two tools for the same job means you've mastered neither and you're paying twice.
• Start at $0. A free ChatGPT or Claude account, an image generator's free tier, and the AI already built into DaVinci Resolve is a complete starter stack. Ship one project on it before spending a dollar.
• Pay only for bottlenecks. Upgrade a tool when its free tier is blocking actual work this week — never because a launch video impressed you.
• One project, then decide. Every trial gets tied to a real project with a real end date. When the project wraps, the tool either proved itself or gets cancelled that day.
• Read the data policy before you upload. If you can't find a plain answer to "will my uploads train your models," assume the answer is yes and act accordingly.
• Re-audit quarterly. Every three months, open your statement and cancel anything that didn't touch the last project. No sentiment, no "but I might."

The trial rule deserves emphasis because it's where discipline actually lives. Most people trial a tool by poking at it for an evening — that tells you nothing except whether the onboarding is slick. Trial it by giving it a job: board one scene, transcribe one interview, draft one synopsis. A tool that survives contact with a real deadline has told you something. A tool you only ever "explore" is entertainment, and you should budget it as entertainment.

Where your script goes when you hit upload

Now the part almost nobody reads before clicking agree. Many AI services reserve the right to use what you upload to improve their models — and the terms vary wildly between free and paid tiers, and between consumer and business accounts. Some let you opt out; some make retention hard to pin down. The details change too often for me to summarize them here, which is exactly the point: check the current policy of the specific tool, on the specific tier you're using, before you upload anything you'd fight to protect.

For filmmakers, the sharpest edge is the unproduced script. It's your most valuable asset precisely because nobody else has it. Before you paste a full draft into any service for feedback, know where it goes. Sometimes the sane workaround costs you nothing: get notes on a detailed treatment or a few scenes instead of the whole draft, or use a tier that contractually excludes your data from training. This is unglamorous homework, and it's also the difference between using these tools and being used by them. The same caution goes double for anything you don't own outright — a script you optioned, a producer's confidential coverage, footage with someone else's face in it.

Build the small stack, work it hard, and upgrade only when the work demands it. Next chapter we deal with the questions that money can't answer — who owns generated material, whose consent you need, and what you owe your collaborators and your audience when AI touches the film.