Pricing
Every plan writes the prompts, wires the canvas, prices the set and renders it in order. Every plan gets every model. What changes higher up is how much runs without you in the room — and how much of the office inherits it.
Images cost by resolution and video costs by the second. Prompts cost roughly what they cost us to serve. The pre-flight check that stops a wasted render is free, and a failed call refunds itself.
One credit is one US cent, and everything is priced at the same markup over what it costs us to run. GPT Image 2 is billed by quality and by shape rather than by pixels, so its numbers move with the frame — the canvas shows you the exact figure on the button before you press it, and updates it the moment you change the ratio.
| Action | Credits | What that means |
|---|---|---|
| Nano Banana 2 · 1K | 15 | Working resolution, fine for a review |
| Nano Banana 2 · 2K | 22 | Presentation resolution |
| Nano Banana 2 · 4K | 33 | Print, or a hero on a board |
| Nano Banana Pro · 1K–2K | 29 | The hero shot |
| Nano Banana Pro · 4K | 53 | Print at full size |
| GPT Image 2 · low | 3 | Cheap enough to iterate freely |
| GPT Image 2 · medium | 9–12 | A wide or tall frame costs less than a square one |
| GPT Image 2 · high | 36–46 | Same again — shape changes the price |
| Upscale to 4K | 33 | Resolution only — nothing reinterpreted |
| Writing a prompt | 0.5 | About what it costs to serve |
| Asking Nomad anything | 0.5 | Same rate, no separate chat plan |
| Pre-flight check | 0 | Our quality control, not your action |
| A failed call | 0 | Refunded automatically |
Every engine bills per second rather than by resolution, so the length control replaces the resolution one and the price follows it. The spread is wide on purpose: the cheapest clip on the canvas is a seventieth of the dearest, which is what makes drafting a move affordable.
| Engine | Credits | What that means |
|---|---|---|
| MiniMax H3 · 768P · 4s | 70 | The cheapest video here. Where a move should be found |
| MiniMax H3 · 768P · 15s | 263 | Its longest take |
| MiniMax H3 · 2K · 8s | 228 | Finishing the move you kept |
| Omni Flash · 4s–10s | 90–220 | 720p, and the only engine needing no relay |
| LTX Pro · 1080p · 6s | 197 | Fidelity, with sound generated alongside |
| LTX Fast · 4K · 10s | 810 | When the frame has to be big |
| FLUX 3 · HD · 20s | 745 | Twenty seconds, up to five shots, one generation |
| FLUX 3 · FHD · 20s | 1,270 | The dearest single act on the canvas |
| A film in one clip | one clip | Several shots cut inside a single generation — charged once, not per shot |
| A sixty-second reel | ~1,240 | Twelve five-second cuts on MiniMax 768P, plus the twelve storyboard stills — most of a Starter month |
Because the model is not the product. Anyone can hand you Nano Banana Pro; what takes the afternoon out of your week is Nomad reading the canvas, asking the one question it cannot answer itself, writing four hundred words of prompt properly, and queueing eighteen renders in the right order. Every plan gets all three models from the first day.
What you buy higher up is scale of automation — how much runs unattended, how much of it the whole office inherits, and how far it reaches into the tools you already use.
A still costs tens of credits; twenty seconds of FHD video costs over a thousand. That gap is not a pricing decision, it is what the providers charge, and it changes how the tool should be used: find the move at 768P for 70 credits, watch it, and spend the real money once on the take you kept. The canvas is built for that order — the engine sits on the node, so switching from draft to finish is one press.
A normal working day is around twenty-five prompts and a handful of renders. On Starter that is comfortable — 133 images a month is about six a working day. A studio producing full sets weekly lands on Professional. If you are somewhere between, start lower — moving up mid-month is one click and takes effect immediately.
Storage is the other half of the decision. Five gigabytes holds a few thousand reference photographs and a month of renders; ten holds a practice's worth. It costs us very little, so the limits are there to bound the extreme case rather than to recover a cost — if you are close to yours and everything else fits, ask and we will raise it.
Video changes that arithmetic, so it is worth saying plainly: Starter's 2,000 credits is 133 images at 1K, or fourteen drafted eight-second camera moves, or a single sixty-second reel — and those are alternatives, not additions. A practice making reels weekly belongs on Professional.
Uploads are different: they cost no credits at all, only space. That is what the storage allowance is for — dropping three hundred reference photographs into a project to think with should not cost you anything, and it does not.
Monthly credits do not roll over. Top-ups do not expire at all, which is the way round we would want it if we were buying.
There is no free tier. The canvas replica is open to walk through without an account, and it costs you nothing to see exactly how the tool behaves before you decide.
Not with generation, no. Running a render costs real money the moment it happens, and a free tier would either be too small to judge anything or too expensive for us to keep honest.
Instead the canvas replica is open to anyone — the real workspace, the real panels, the real sequence, with the renders already in it. You can see precisely how it behaves before paying anything.
Monthly credits reset at the start of each period. Top-up credits never expire and are spent only after the monthly ones run out, so you cannot accidentally burn the ones you bought.
Nothing breaks. The canvas stays open, everything you made stays there, and prompts you have already written keep working. Buy a top-up or move up a plan and rendering resumes immediately.
All of them, on every plan. Three for stills — Nano Banana 2, Nano Banana Pro and GPT Image 2 — and four for video: MiniMax H3, LTX 2.5, FLUX 3 and Omni Flash. You choose per node rather than per project, so a single set can mix them: the fast one for iteration, the Pro for the hero shot, 768P to find a camera move and FLUX for the finished film.
We do not gate models by price. The model is not the product; what saves you the afternoon is Nomad writing the prompt — in the grammar that particular model was trained on — and running the set.
Because it is, at the provider. A four-second clip is roughly two 4K stills; twenty seconds at full resolution is closer to forty. We apply the same markup to video as to everything else, so the gap you see is the real one.
The practical answer is to draft. Find the move on MiniMax at 768P for 70 credits, keep the one that works, and spend properly once. The engine is a setting on the node, so that switch is a single press.
For one engine, briefly. Some video APIs do not accept requests from a browser at all — a rule the browser enforces, not a choice of ours — so Nomad falls back to a small relay you run locally when that happens. It is one file and one command, and Omni Flash needs none of it.
No. Your images go to the model provider you chose for that node and nowhere else. The inspector shows exactly what was sent on every request — the prompt, the attachments, the model and the resolution.
On Studio, yes — that plan exists partly for offices that need it, with data handling agreed in writing before you start. On the other plans, treat it as you would any cloud tool: the providers' terms apply, and we will point you at them rather than paraphrase.
Any time, from your account, without a call. You keep access until the end of the period you have paid for, and your projects stay downloadable afterwards.
Not yet, and you do not need one. Screenshot your viewport and drop it on the canvas — anything you can see, Nomad can read. Plugins are on the way for people who would rather not leave the modeller.