Almost every AI video tool in the market launched with a subscription plan. Pro, Premium, Studio, Enterprise tiers. $20 a month, $50 a month, $200 a month. The pattern is so universal that nobody pitches a deck without it.

It's also wrong for AI video.

We tried subscriptions for nine months. We killed them. Cleom now runs entirely on pay-as-you-go credit packs. This post is why, and what changed when we made the switch.

What's wrong with subscriptions for AI video

The case for subscriptions in traditional SaaS is well-understood. Usage is consistent. Costs are mostly fixed (engineering, hosting). The customer derives roughly the same value every month. Bill them the same every month. Everyone's incentives align.

AI video breaks every one of these assumptions.

Usage is bursty

An AI video user doesn't generate uniformly. They have a project — a brand spot, a YouTube video, a presentation reel — and they generate intensely for a few days. Then they don't touch the tool for two weeks. Then another project, another burst. Real usage looks like this:

  • Day 1-3: project deadline, 40+ generations, heavy use
  • Day 4-14: nothing
  • Day 15-17: new project, another burst
  • Day 18-30: nothing

On a subscription, this user is paying every day for a tool they use eight days a month. They feel like they're being overcharged. They are.

Costs are usage-driven

Unlike traditional SaaS, AI video has real per-unit costs. Every generation burns GPU time. Models, especially video models, are expensive to run. The marginal cost of a single 4K 30-second generation is a real number — not zero.

This means when a subscription user generates 40 videos in three days, they're costing the provider more than their monthly fee. The provider has to either price the subscription high enough to cover the heaviest users (which screws everyone else) or eat the loss (which doesn't scale).

The pricing signals are wrong

Subscription users have a perverse incentive — they want to "get their money's worth." They generate things they don't need just to maximize the perceived value of the plan. This burns GPU on outputs that go straight to the trash. The tool gets blamed for being "expensive per useful output," even though the cost per generation is the same.

Pay-as-you-go flips the incentive. Users only spend credits on outputs they actually want. Generations that ship are the only ones that cost money. The economics align.

What we replaced subscriptions with

The model we landed on at Cleom is credit packs, with no recurring billing. You buy credits, you use them when you need them, they don't expire. If you generate nothing for three months, you spend nothing. If you produce a film in a week, you buy a bigger pack.

The pack tiers map to project sizes:

  • Starter ($10) — try the platform, generate a few clips, see if it works for you
  • Pro ($50) — produce a small project, a few minutes of finished video
  • Studio ($200) — one full production, like a brand spot or a music video
  • Enterprise ($500+) — agency-scale work, multiple campaigns

The bigger packs include bonus credits — buy in volume, pay slightly less per credit. This rewards heavy users without forcing them onto a subscription that ticks during weeks they're not working.

What changed when we switched

Three things happened that we didn't fully expect when we made the switch.

1. Conversion went up

The big psychological barrier to subscriptions is "am I going to use this enough?" People know themselves. They know they're not going to use the tool every day. So they don't subscribe.

Credit packs remove that barrier. The math becomes simple: "I have a $50 project to make. Does this tool let me make it for $50?" If yes, they buy the pack. There's no commitment, no recurring drag, no annual contract.

Our trial-to-paid conversion roughly doubled within the first month of the new model.

2. Customer relationships improved

Subscription customer support is almost entirely about cancellations and refund requests. "I forgot to cancel, can I get my money back?" "I didn't use it last month, please refund." Half of support tickets were these.

Credit packs eliminate this category entirely. There's nothing to cancel. There's no monthly forgetting. Customers either have credits or they don't. Support shifted from billing disputes to actual product help.

3. Revenue stayed flat (then grew)

The honest worry going in was revenue. Subscriptions are predictable. Credit packs are not. Our existing subscribers were monthly recurring revenue we'd be killing.

What actually happened: revenue stayed flat for three weeks while existing subscribers used up their credit grant (we converted everyone's remaining subscription value into credits, no penalty). Then it started growing — because the new customers who wouldn't subscribe were now buying packs.

By month two, revenue was higher than it had been on subscriptions. By month six, it was 40% higher.

Subscriptions priced for the average user. Packs let each user price themselves.

Where subscriptions still make sense

To be fair, we still talk to teams that prefer a subscription. Almost all of them are agencies or studios with predictable monthly throughput. They want a fixed line item in their budget. They use the tool every day. For them, the unpredictability of pay-as-you-go is a feature loss, not a feature gain.

Our Enterprise offering accommodates this with custom contracts — committed credit allocations on a quarterly or annual basis, at preferred rates. It looks like a subscription on the surface, but it's a commercial arrangement, not a default pricing model. The default for everyone else is packs.

Conclusion

Subscriptions made sense when SaaS was about consistent, low-marginal-cost software. AI video isn't that. It's high-marginal-cost, bursty, project-driven. The pricing model has to reflect that.

Pay-as-you-go credits aren't a gimmick or a positioning play. They're a better fit for how the work actually happens. If you're paying for an AI video subscription right now and using it less than you'd like, that's not a discipline problem. It's a pricing-model problem.

You shouldn't be subscribing to AI video. You should be paying for what you generate. Nothing more.