Search for a free AI video generator on Mac and you get four different products wearing one word. They fail in different ways, and the difference only shows up after you have committed a weekend to one.
1. Free tier of a hosted service
A web app with a credit allowance. Genuinely free until the credits run out, which — at the resolutions anyone actually ships — is usually a handful of clips. The generation happens on their hardware, so your footage goes to their servers, and the output often carries a watermark or a licence restriction on commercial use.
Fine for finding out whether the technology does what you hoped. Not a production pipeline.
2. Free app, paid engine
A Mac app you download for nothing that calls a paid API behind the scenes. The download is free; the generation is metered. This is the most common shape and the most commonly misread, because the price appears after installation rather than before it.
Worth checking before you invest time: whose key is it, and what is the per-second rate.
3. Open weights you run yourself
The model runs on your machine, nothing is uploaded, no meter exists. On Apple Silicon this is real and it works — unified memory means the GPU addresses the whole machine, so a 64GB Mac holds a pipeline a 24GB discrete card would have to offload.
What it costs instead is patience and setup. A hero clip is tens of minutes, not seconds: one of our real renders at 1080×1920, 97 frames, took 1,830 seconds. You are also managing a Python environment and a model port rather than clicking a button.
4. Machines you already own, run as a farm
The same open weights, but across several Macs pulling from one queue. This is what we do, and it is the only one of the four with $0 per render in the literal sense — no meter, no credits, no per-second rate, and the footage never leaves your network.
The gain is throughput, not speed. Four Macs give roughly 4× on a batch because the jobs are independent. They give nothing on a single clip: one render cannot be split across machines, and the reason is worth understanding before you plan around it.
Which one you want
Trying the idea out: a hosted free tier, and stop there until you know it works for your material.
Occasional clips, no patience for setup: a free app with a paid engine, eyes open about the rate.
Footage that cannot leave the building, or volume that makes a meter uncomfortable: run the weights yourself. If the work is batch-shaped — forty clips by morning rather than one clip now — do it across the Macs already sitting idle after 5pm.
The part nobody puts in the comparison
A meter changes behaviour more than it changes budget. When each attempt costs something, people stop exploring: they batch decisions, skip the variation they were curious about, and settle earlier than they should. Removing the meter is worth more to the work than the money it saves, and that is the actual argument for running it yourself.
Setting that up starts here, or read what to expect from a Mac first.