What's actually free in AI video (and what nobody can give away)

By a wide margin, the most common thing people search for in this category is a free AI video generator. Not the best one. Not the fastest. Free.

That is a completely reasonable thing to want, and it deserves a straight answer rather than a landing page that says "free" in large type and reveals the catch after you have signed up. So here is the actual economics, including ours.

Why generating video cannot be free

Every generated clip is GPU time. A few seconds of video is minutes of expensive hardware running flat out, and somebody is paying for that hardware by the second — it is a real, per-clip cost that does not go away with scale the way storage or bandwidth does.

This is the difference between AI video and almost every other free web tool you have used. A photo editor costs its maker essentially nothing per edit. A video generator costs its maker real money every time you press the button, including the four times you press it because the first three rolls were wrong.

So when a service offers unlimited free generation, it is not being generous. It has chosen one of the options below.

The five ways "free" is actually paid for

Trial credits. You get a fixed amount, framed as free. It is a sample, and it runs out. Perfectly honest as long as the number is stated up front — the problem is when it is not.

A watermark. You can generate, but the output carries branding. Free for personal use, unusable for anything client-facing or commercial, which is the point.

A resolution or length cap. Free output is 480p, or five seconds, or both. Fine for testing whether a prompt works, not for anything you would publish.

The queue. Free generations go last, behind everyone paying. At busy times "free" means "tomorrow".

Your data. The one to actually watch. If nothing else is being charged, read what rights you are granting over what you upload and generate — whether it trains future models, and whether you own the output commercially. This is where "free" gets genuinely expensive.

Most services combine two or three. None of them skip all five, because none of them can.

What is genuinely free here, and what is not

We do the same arithmetic as everyone else, so it is worth being specific about which side of the line each thing falls on.

Free, with no credit card and no catch:

• The whole editor. Multi-lane timeline, trimming, transitions, keyframes, colour, blur, masks, text and titles.
• Captions, including word-level.
• The audio mixer, including ducking.
• Camera moves on stills — the Ken Burns push and drift that a great many videos are mostly made of.
• Unlimited projects.
• Exports at 4K with no watermark — ten per month on the free tier.
• 1 GB of asset storage.
• Tokens to start with. A new account is credited enough to run a video generation and a few images, so you can see what the models actually do before deciding anything. No card, nothing to claim.

Not free, after that:

• Generation. Video, images, voices, music. This is the GPU cost, and we pay it to the model providers per request.

Once the starting tokens are gone the way forward is tokens, topped up when you want them — there is no monthly allowance on the free tier and no subscription required to buy a pack. See pricing for the current numbers.

So we do the same thing as the trial-credit tools on that first bit, and we are not going to pretend otherwise. Where we differ is the rest of it: what you will not find here is generation quietly capped at 480p with a logo in the corner, because we would rather charge for the expensive thing and give away everything around it — the editor, the captions, the mix, and 4K exports with no watermark, for as long as you like.

How to spend the least, whatever you use

The single biggest lever is not which service you pick. It is how many generations you burn per finished video.

Explore with stills. Images cost a fraction of video. Iterate on a still until the look is right, then animate the one you kept. People who find AI video expensive are usually burning video generations on look development.

Use camera moves for most shots. A slow push across a still costs nothing — it is an editing effect, not a generation. A ten-minute video built from forty moving stills and six generated clips costs a small fraction of one built from forty generated clips, and for most formats nobody can tell which shots were which.

Generate short. Clips degrade the longer they run, so a long generation is both more expensive and more likely to be unusable.

Fix instead of re-rolling. If a clip is nearly right, upscaling or frame interpolation is far cheaper than another roll — and it keeps the take you already liked, which a re-roll cannot give back.

Use the cheap tier for the exploring pass. Fast model variants exist precisely so the throwaway attempts and the final shot do not cost the same.

Five questions worth asking any free tool

1. How many free generations, exactly, and then what?
2. Is the output watermarked?
3. What resolution and length, on the free tier specifically?
4. Do I own the output commercially?
5. Is what I upload or generate used to train anything?

The answers are usually in the terms rather than on the pricing page, and a service that makes them hard to find has told you something.

We would rather you asked us those five questions than found out later — which is the whole reason this post exists, given "free" is what most people searching for this are typing.