Free video is rarer than free images because a clip is hundreds of frames of GPU time rather than one. It exists here for the same reason free images do: two of our video models run on hardware we already own, so they can be paid for with the green wallet that refills itself instead of with money. The allowance is small and it carries one hard clamp. Here is the arithmetic first, then the instructions.
An OpenModels account carries two wallets and they do not convert into each other. Green credits (⚡) refill on a clock and can only pay for models that run on hardware we own. Gold credits (✦) are bought with money, or granted by a paid plan, and pay for everything else — the hosted endpoints from Google, ByteDance, Kling and the rest, which cost us real dollars every time they are called. Signing up grants zero gold.
So the free video lane is exactly the set of video models on our own GPUs. There are two of them, and one has four separate modes:
Which plan: Free plan, signed in: ⚡20 a day, refilling continuously toward that cap rather than landing as a lump at midnight, and every video render clamped to a 480 px short side. Every paid plan raises the daily cap and removes the clamp entirely.
LTX is billed per started five-second block, not per second, and the block price is quoted at a 720p short side. Smaller renders cost proportionally less GPU time, so the 480 px size the free plan is clamped to carries a 0.75 multiplier. That is the whole calculation: block price, times blocks, times 0.75.
| Clip length | Five-second blocks | Cost from | Clips in a ⚡20 day |
|---|---|---|---|
| 4 seconds | 1 | ⚡4.5 | about 4 |
| 8 seconds | 2 | ⚡9 | about 2 |
| 16 seconds | 4 | ⚡18 | 1 |
MiniMax H3 Studio is priced by the second instead, from ⚡2 a second, and takes any length from 1 to 15 seconds. A five-second H3 clip and an eight-second LTX clip cost roughly the same. Which one is the better use of the day depends entirely on whether you need reference conditioning or a start frame.
What it costs: Every number above is a floor, not a quote. Prices on our own hardware are set at runtime, can carry a demand multiplier when the render queue is deep, and when the local lane is unavailable the same model is served over a paid API and billed in gold instead. The composer quotes the real price on the Generate button before you press it.
The clamp sets the short side of the output. A 16:9 clip comes back at roughly 854 by 480; a 9:16 vertical comes back at 480 by 854. It is applied on the server before the price is worked out, deliberately: quoting 1080p and then rendering 480p would have billed people for pixels they never received, and it did, once.
What the clamp does not do is change the model, the prompt handling, the length options or the aspect ratios. LTX offers 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16 on every plan. A 480 px vertical clip is a perfectly usable social draft — it is a poor master. Treat the free lane as the place you find out whether a shot works before spending a paid render on it at 1080p.
Open the composer — Opens the video composer. Nothing runs until you pick a model and press Generate.
Being straight about this is more useful than a longer feature list, and it is the whole point of this library.
| Limit | On the free plan | What lifts it |
|---|---|---|
| Daily green credits | ⚡20, refilling hourly | Starter 100, Creator 200, Pro 400, Ultra 2,000 |
| Output size | 480 px short side | Any paid plan removes the clamp |
| Hosted video models | Not reachable — green cannot buy them | Starter and up, paid in gold |
| Parallel jobs | 2 at a time | Starter 3, Creator 5, Pro 10 |
| Watermark | Always on | Creator and up can switch it off |
| Storage | 50 MB | Every paid tier raises it |
Watch out: Green credits are never for sale. Buying credits buys gold, which pays for hosted models and queue priority — it does not raise your daily green allowance or lift the 480 px clamp. Only the plan does that.
Credits are not the scarce thing on the own-GPU lane. Time is. There is one GPU behind every own-GPU model on this platform, and it runs one job at a time, so a video render that takes several minutes sits in a line with everybody else's. That is the honest trade a free tier on our own hardware makes: the electricity is cheap, the machine is not duplicable.
The composer shows the real position rather than a spinner. Every queued job carries its place in the line and the board's current wait estimate, and every queued job can be boosted into a faster lane — Fast doubles the cost of the run, Express quadruples it, and the multiplier is snapshotted at submit so a busy minute later cannot change what you agreed to. Congestion also raises those multipliers, which is the mechanism that keeps the fast lanes fast.
No. The green wallet refills toward its daily cap every hour, for as long as the account exists. There is no card required and nothing expires into a paywall.
At the clamped 480 px size, about one 16-second LTX clip, two 8-second ones, or four 4-second ones. Own-GPU prices are set at runtime and can rise when the render queue is deep, so treat those as approximate.
No. Those are hosted models billed per call by their provider, so they are paid in gold credits and open from the Starter plan. The picker shows them with the plan that opens them rather than hiding them.
The free plan clamps the short side to 480 px, and the clamp is applied before pricing so the quote matches what you got. Any paid plan removes it and the same models render up to 1080p.
Yes. Output on the Free and Starter plans is watermarked. Creator and higher plans carry a switch, and it is re-read on every save rather than at signup.