Most number posts are about growth. Ours is about the product: the nine figures that decide what you get, what it costs and where the edges are.
We picked them because every one is something you can check yourself — in the model picker, on the pricing page, or in your own balance.
Why these nine
Usage and revenue figures are interesting to us and not very useful to you. What you need to know before paying for a tool is how it behaves: which models it uses, what a render costs, what the free plan really includes, and where the limits are.
Numbers you can check beat numbers you have to trust.
— Rui, annual note
The models
4 — image models: Nano Banana 2 Lite, Nano Banana 2 and Nano Banana Pro from Google, and GPT-5 Image Mini from OpenAI
3 — video models: Seedance 2.0 Fast and Seedance 2.0 from ByteDance, and Hailuo 2.3 from MiniMax
0 — models we train or host — every render is a call to a third-party model through OpenRouter
The credits
1 — credit for a Draft image, the cheapest render on ockeeper
50 — free credits every month, reset on the 1st (UTC)
~17% — what a yearly plan saves against twelve monthly payments
Credits arrive monthly on every plan, yearly ones included, and unused credits expire at the end of the month. A render that fails, or is refused by the model's safety filter, costs nothing.
The limits
4096 px — the largest output, on GPT-5 Image Mini and Nano Banana Pro
15 s — the longest video clip, at up to 1080p, on paid plans
1 — business day — how quickly we respond when you write to us
What the numbers leave out
The free plan is Draft only: Nano Banana 2 Lite, up to 1024 px, with a visible ockeeper mark on every image — and still licensed for commercial use. Paid plans unlock the other models, video, and images with no mark.
Every image gets a commercial licence certificate number. We sell subscriptions only, with no top-up packs, and purchases are not refunded; your statutory rights in the EU and UK are unaffected.
What stays the same
When the product changes, these numbers change with it. The pricing page and the model picker are always the current version; this post is a snapshot.
We do not train models on your renders — we do not train models at all.
Thank you for reading the small print with us. If any number here does not match what you see in the studio, tell us — that is a bug.