Credit Economics: The Resolution Ladder and Budgeting a Production
Most wasted generation budget isn't bad prompting — it's defaulting to maximum resolution on takes you were never sure about.
Budget discipline is a directing skill
Every lesson in this track so far has been about getting a better result out of a single generation — better keyframes, better shotlists, cleaner continuity, fewer defects. This lesson is about something different: making sure you can afford to apply all of it across a real production, not just a single lucky shot.
Credit economics is not an afterthought bolted onto the craft. It's the same kind of discipline as everything else you've learned — a set of rules that, followed consistently, turn an unpredictable cost into a manageable one. Directors who ignore it don't fail because they lack technique. They fail because they burn their budget testing things cheaply and then have nothing left for the shots that actually needed the spend.
The resolution ladder: the single biggest cost lever
If there is one habit that separates a director who runs out of budget from one who doesn't, it's this: they never default straight to maximum resolution.
Think of resolution as a three-rung ladder. The bottom rung is a low-resolution test tier — cheap, fast, good enough to tell you whether a prompt, a motion phrase, or a keyframe pairing is even going to work. The middle rung is your everyday default — the resolution you actually shoot most of a production at, upscaling in post later if a specific shot needs more fidelity. The top rung is reserved for shots you are already certain about: hero beats, final deliverables, anything where you've already validated the approach and just need the final-quality pass.
The mistake that burns the most budget in this craft is skipping straight to the top rung on every take, including throwaway tests. If you're not sure a prompt will even resolve the motion you're describing, testing it at maximum resolution is paying premium price for information you could have bought cheap. Reserve the top tier for confirmation, not discovery.
Cost facts worth internalizing
A few realistic, illustrative figures make the ladder concrete rather than abstract. A full roughly-15-second clip at the mid resolution tier runs somewhere in the range of 60-70 credits — that's your everyday production cost per shot. A single high-resolution identity-anchored still frame, by contrast, is very cheap — a small fraction of a credit — cheap enough that you can generate many candidate stills freely when you're locking a look or a reference frame. And a generation that gets content-flagged or blocked is typically not charged at all, which changes how you should think about risk: a risky exploratory prompt costs you time waiting for the block, not credits. That means the correct sequencing is to front-load your riskiest exploration — the prompts you're least sure will clear moderation or resolve cleanly — before you've locked in a safe, validated phrasing, rather than discovering the risk late when you can least afford the detour.
Budgeting discipline: rerolls are a normal cost, not a failure
New directors treat every generation that isn't perfect on the first try as a mistake to be embarrassed about. That's the wrong frame, and it leads to bad budgeting, because it means you never planned for the rerolls you were always going to need.
The professional discipline is to budget 5-8 rerolls per genuinely important hero shot as a normal cost of the craft. "Generate many, pick the best, and recombine" is the real workflow — not a fallback for when things go wrong, but the intended process. If you budget for this up front, a shot that takes six takes to land doesn't blow your production budget; it's exactly what you planned for. If you don't budget for it, the same six takes feel like a crisis, and you'll cut corners on the shot that mattered most to save credits for shots that didn't need them.
A worked budget example
Put the ladder and the reroll discipline together and the arithmetic gets concrete fast. Say a production needs twelve shots: two are hero beats (the reveal and the closing shot), and the other ten are standard coverage. Under the ladder discipline, none of the twelve get tested at the top tier — every shot's prompt gets a cheap low-tier pass first to confirm the motion resolves the way you intend, and only the two hero shots graduate to the top tier once validated.
Budget the reroll discipline on top of that: 5-8 rerolls on each hero shot, maybe 2-3 on standard coverage shots where the stakes and the uncertainty are both lower. A director who skips this arithmetic and assumes "one clean take per shot" will blow through their credit balance the moment the first hero shot needs its sixth take — and will likely respond by cutting quality on the shot that mattered most, because the budget wasn't planned to absorb it. A director who budgets the reroll count up front treats the sixth take on the hero shot as exactly what was planned, not an emergency.
Flat-rate windows: only pay off pre-validated
Some providers sell a flat-rate, unlimited-generation window — a 24-hour pass at a fixed price being the common shape. These can be extremely good value, but only under one condition: your prompt, reference, and skeleton approach has to already be validated before the window opens.
The economics work like this. If you walk into the window with every storyboard beat, every prompt spec, and every reference image already locked — the only thing left to do is fire the batch — then the flat price buys you effectively unlimited iteration on a known-good plan, which can be worth many times the fixed cost. If instead you walk into the window planning to figure out your prompts and references live, most of the paid time evaporates into the same trial-and-error you'd otherwise do for free at the low resolution tier — except now you're doing it against a clock, under time pressure, at whatever resolution the window defaults to. Pre-building before the window opens is not optional polish; it's the entire mechanism by which the flat rate pays off.
Concurrency: a load problem, not a prompt problem
One operational note that trips up directors trying to use a flat-rate window efficiently: firing many generation requests at once against the same provider can trigger transient server errors under load. It is tempting, when a request fails, to assume the prompt itself was the problem and start rewriting it. That's usually the wrong diagnosis. The correct engineering response to a burst of transient failures under high concurrency is to serialize the requests and retry the ones that failed — treating it as a load issue, not a content issue. Chasing prompt rewrites for a batch of requests that failed because you fired forty of them at once wastes time solving a problem you don't actually have.
Build it: a production cost estimator
You're going to build the tool that turns a shot list into a real budget: sum estimated credits across every shot, accounting for resolution tier and a reroll budget per shot, compare that total against an optional flat-rate window price, and flag when a shot violates the resolution-ladder discipline — a test shot requested at the maximum resolution tier.