What unlimited generation offers actually cap
Time-boxed unlimited promotions cap resolution, model subset and queue before they cap volume. Six levers to check, and how to find the real ceiling.
The day counts give it away. Through 2026 the AI video category has run a rolling series of "unlimited" promotions, and Higgsfield published its own as blog posts with numbers like 33 days, 14 days and 11 days attached (Higgsfield's offer posts). Nobody arrives at 33 days by accident. Those numbers are chosen because they read as specific rather than promotional, and because a month that doesn't round is a month you're more likely to act on this week. The day count is the part of the offer everyone repeats. It is also the least important cap in it.
An unlimited generation offer is a bundle of constraints wearing one word. Volume is genuinely uncapped inside the box; the box itself has six walls. Here's how they're built, in roughly the order they cost you something.
The six walls
1. The time box. The obvious one. It runs from your activation date, not from when the offer was announced, and it typically ends whether or not you generated anything. Odd day counts also make expiry harder to track mentally — you know when a month ends, you don't know when 33 days ends.
2. The model subset. "Unlimited" is almost never unlimited across a catalog. Higgsfield's offers were built around a single named model — Seedance 2.5 in the 2026 run, with Seedance 2.0 sitting at a different tier — and everything else on the platform stayed metered. That's the wall that costs the most and gets checked the least, because a model subset of one is a very different product from a catalog. For scale, Versely's catalog carries 296 models: 146 video, 108 image, 22 speech and audio, 19 lipsync and avatar, and one video upscaler. An offer covering one of those is covering roughly a third of one percent of what you'd otherwise reach for.
3. The resolution tier. This is where the actual money sits. Higgsfield's published terms put 720p on the newer model and 1080p on the older one, and a separate 4K offer ran on a much shorter window — an 11-day one against the 33-day headline. When the high-resolution version of the same promotion is a third of the length, the resolution tier is telling you where the cost actually is.
4. The toggle. Higgsfield's offer posts are explicit that an "Unlimited" toggle has to be on. If it's off, generations draw credits from your balance exactly as they would have without the promotion. That is a real, documented failure mode: you believe you're inside the offer, the UI says otherwise, and the meter runs.
5. The plan and billing cadence. The headline number is the best case. In the 2026 Higgsfield run the actual duration landed somewhere between 22 and 33 days depending on which plan you were on and whether you billed monthly or annually. The number in the title belongs to one row of a matrix.
6. Queue priority and concurrency. Unlimited volume and unlimited throughput are different promises. If everyone on the offer is sharing a pool, the practical ceiling is how many jobs you can have in flight and how long each one waits — which is a cap even when no document says "cap".
Why the resolution tier is the expensive wall
People treat resolution as a quality preference. On a metered platform it's a rate lookup, and the spread is much wider than the intuition suggests.
Across the 52 resolution-banded per-second video models in Versely's catalog, moving from a model's cheapest band to its dearest multiplies the per-second rate by anywhere from 1.29× to 19.57×. That is not a typo, and it is not one model being weird — it's the honest range across a real catalog. The 4K-versus-1080p cost page works the arithmetic, but the shape is simple: resolution doesn't add a surcharge, it selects a different per-second rate, and that rate then multiplies every second of the clip.
A concrete example from the catalog. LTX 2.3 Text to Video Fast charges 4 credits per second at SD/HD and 16 credits per second at 4K. A ten-second clip is 40 credits at the lower band and 160 at the higher one. Same model, same prompt, same length.
Now read an unlimited offer against that. Capping output at 720p while leaving volume uncapped means the expensive dimension is exactly the one you can't touch. Volume was never the costly axis. Resolution and duration are, and both are usually walled.
| Wall | Where you'll find it | What it actually costs you |
|---|---|---|
| Time box | Headline | Nothing, if you use it |
| Model subset | Body copy, one model name | The rest of the catalog stays metered |
| Resolution tier | Fine print or a tier table | The single largest rate lever there is |
| Toggle | A settings screen, sometimes a footnote | Silent credit burn when it's off |
| Plan / cadence | A matrix, not a sentence | Up to a third of the headline duration |
| Queue and concurrency | Usually nowhere | Throughput, on the days it's busiest |
Read the offer in this order
Seven questions, in the order that resolves the value fastest. If the first three come back badly, the rest doesn't matter.
- Which exact models? Not "video models" — the model names. If the answer is one name, price the offer as access to one model.
- What resolution ceiling, and what is that model's own ceiling? The gap between those two numbers is the real discount you're not getting.
- What duration ceiling per generation? An uncapped count of five-second clips is not an uncapped minute count.
- Does anything have to stay switched on? If yes, find the switch before you generate, not after.
- What's my actual day count on my plan and billing cadence? Read the matrix row, not the headline.
- What happens to work in progress at expiry? Queued jobs, saved drafts, and anything mid-render.
- Is there a concurrency or rate limit? If it isn't published, assume there is one and test it on day one with a batch rather than on day 20 with a deadline.
Then do the arithmetic that actually settles it: take the work you genuinely intend to ship in that window, price it at the resolution you actually need, and compare that to the offer. Most people compare against the work they imagine doing at unlimited volume, which is not a real number. Estimating the credit cost before you dispatch a batch covers doing that honestly.
What the metered alternative looks like
Versely doesn't run an unlimited offer, and it's worth being direct about the trade rather than pretending metering is a feature. Every generation costs credits. The floor across the whole catalog is 1 credit; nothing generates for zero. Plans and credit costs are published on pricing, and what a credit buys breaks down the per-generation side.
What you get in exchange for the meter is that no wall moves. Every plan reaches the same 296 models, the same editor and the same agent — the only thing a bigger plan changes is how many credits arrive and how often. There's no model subset, no resolution tier gated behind a plan, and no toggle that silently changes what a generation costs. The published workflow recipes show what finished multi-scene work actually totals, from 60 credits for the viral panel reaction reel up to 1,800 for the largest recipe at preview resolution — real numbers rather than a promise of "as much as you want" with the expensive axis pinned.
One genuine zero-credit path exists and it's narrow enough to state precisely: the editor renders preview passes at 480p at no credit cost, subject to a short per-user cooldown, and charges once for the final export regardless of how many clips are on the timeline. That's iteration being free, not generation being free. The previews-and-export cost page has the formula.
If you do take an unlimited offer, the right way to use it is as a testing window rather than a production window: run your model comparisons, your prompt variants and your reroll experiments inside the box, and export the finals at the resolution you actually need wherever that's cheapest. Testing is exactly the workload where volume matters and resolution doesn't.
FAQ
Is an unlimited offer ever the cheaper option?
Yes, in one specific shape: high-volume, low-resolution, single-model work over a short, defined period — model bake-offs, prompt tuning, hook testing, anything where you're generating a hundred variants to pick three. That workload is capped by count, which is the one axis the offer actually frees. Finished client deliverables at full resolution are capped by the walls the offer leaves standing.
Why do these offers use odd day counts?
Specificity reads as a real, engineered limit rather than a marketing round number, and a non-round window is harder to plan around, which pushes activation forward. It also lets the same offer be re-run at 11, 14, 22, 30 and 33 days without any of them looking like a repeat of the last one.
If output is capped at 720p, can I just upscale afterwards?
You can, and it's a different operation with a different result. An upscale pass reconstructs detail from what's in the frame; it doesn't retroactively give the generator more pixels to plan the shot with. For a scroll-past social cut that's often fine. For anything where fine detail carries the message — text on packaging, fabric, small product features — generating at the resolution you need beats upscaling into it.
Does Versely have an unlimited plan?
No. Every plan is a credit grant, every generation costs credits, and the cheapest job in the catalog is 1 credit. What doesn't vary by plan is capability: the same models, the same editor and the same agent on every tier, with credit volume and billing cadence as the only differences.