Change orders when variants are nearly free

"Just one more version" used to be a request you could price, because making it cost something you could point at. A re-shoot was a day. A re-cut was an afternoon. The cost was visible to both parties, which is why a change-order clause tied to output count worked: the count and the effort moved together.

They no longer move together. Another aspect ratio is minutes. Another eight text variants is a batch. Another take of the same shot is a re-roll. And because the marginal output is close to costless, the request has stopped feeling like a request — which is precisely what makes it unbounded. A trigger that fires on volume now fires late, fires on the wrong things, or never fires at all.

The fix is to move the trigger onto what is still expensive: direction.

One caveat before any of it. There is no public data on how studios have restructured revision terms for cheap generation. No survey, no contract corpus, no study. Practitioner lawyers have converged on a clear minimum clause set for AI work — disclosure, an IP warranty carve-out, a training-data exclusion, and a liability cap tied to the fee paid, with metadata preservation and audit rights in the fuller versions — but none of that touches revisions.

So what follows is reasoning from the economics, not a benchmarked practice. Treat it as a structure to adapt rather than an industry standard to cite.

Why volume triggers misfire now

Three specific failure modes, all of which you have probably lived through.

They fire on the cheap thing. A client asks for the same cut at 1:1 and 9:16 and your clause counts two deliverables. It cost you twenty minutes. Charging for it reads as opportunistic, so you waive it, and the clause loses authority for the request that actually mattered.

They do not fire on the expensive thing. "Same video, but can she be walking instead of standing" is one output, and a revision under a count-based trigger. In production it reopens every shot with the character in it and can burn a day of attempts, because specific character action is exactly where attempts pile up.

They arrive after the money is spent. By the time you have produced enough outputs to trip a count, you have already absorbed the cost of the direction change that caused them. The trigger measures a symptom that appears downstream of the event you wanted to catch.

Four triggers that still work

Each fires on a change of direction, which is the input you actually price. The test for the list: does reversing it invalidate work that is already approved?

1. A locked decision reopens. The cleanest trigger, and the one that covers most real cases. If the scope named the decisions and who signed them — proposition, format, script, look, cast, product treatment, voice, music — then reopening any signed decision is the change order. Scoping a job by decisions rather than deliverables is the structure this depends on; without it, "locked decision" has no referent and the trigger cannot fire.

Price it by depth. Reopening the script is not the same event as reopening the music, and a flat per-round fee prices them identically, which is why flat fees feel wrong to both parties on big changes.

2. A new consistency anchor is required. Anything that introduces a face, a character, a product, a location or a brand element that was not in the approved brief. This is a trigger because the anchor is the expensive part, not the shot. Setting up a reusable character or product, or a reference stack that survives across scenes, is real setup work with real failure modes — four consistency mechanisms and when each fails covers which ones hold under what conditions.

The tell that you are in this territory: the client says "just add the new bottle to the existing shots" and means it as a small ask.

3. The usage changes. New territory, new channel, new duration, new exclusivity, or paid placement where the quote assumed organic. The output does not change — the value does, and so does your exposure. If you sell rights in tiers, this trigger moves a job between them. Usage rights in creator contracts and the usage rights definition cover the tier language; the clause just points at it.

This one catches the request that looks least like a change: "we're also going to run it in Germany."

4. The approval structure changes. A new stakeholder joins after a decision was signed, the named approver is replaced, or feedback starts arriving from a second channel. Not a creative change and not an output change, but it reliably reopens settled decisions, and it is the most common cause of a job going sideways after week two.

Written as a trigger it sounds harsh and reads as reasonable, because the clause is not "we charge for new people." It is "a decision signed by one approver and reopened by another is a change order."

The clause

Paste and adjust. Keep it short — a long change-order clause gets skimmed, and this one needs to be read.

Change orders. The approval decisions listed in the scope are signed by the named owners and lock the work that follows them. Additional outputs produced against signed decisions — further cutdowns, aspect ratios, text variants, or re-selections from material already generated — are included at the volume rate and are not change orders.

A change order applies when: (a) a signed decision is reopened; (b) a face, character, product, location or brand element not in the approved brief is introduced; (c) the usage granted changes in territory, channel, duration, exclusivity or paid placement; or (d) a decision signed by one approver is reopened by another.

Change orders are quoted against the decisions they reopen and agreed in writing before work resumes.

Two things that clause does deliberately. It states what is free first, which is the part the client cares about and is genuinely generous now. And it makes the triggers factual rather than judgemental — each either happened or did not, which keeps the conversation off whether a request was reasonable.

The conversation when a trigger fires

Same principle as any awkward invoice conversation: state it before doing the work.

"Happy to do this — flagging it as a change order before I start so it isn't a surprise. The new presenter reopens the cast decision we signed on the 4th, which means re-generating every shot she's in rather than re-cutting. That's [amount] and about [duration].

If it's useful, the alternative is keeping the current cast and changing the wardrobe and setting instead, which sits inside the signed decision and costs nothing. Happy either way — which do you want?"

The second paragraph is the part that matters. Offering a free path that stays inside the lock proves the trigger is about direction rather than billing, and it quite often gets the client what they actually wanted. Most "one more version" requests are a proxy for a feeling about the work, and a cheaper move sometimes fixes the feeling.

Why cheap iteration makes this easier to hold

There is a version of this policy that is unenforceable, and it is the one where every round costs you money. At that point you are choosing between an unhappy client and an unprofitable job, and the clause is theatre.

The economics are different when exploration is cheap and only the confirmed output is charged. The editor works that way: one timeline, preview: true renders a free 480p pass carrying a short per-user cooldown, and the export charge lands once on the version you confirm regardless of clip count — previews and the final export has the detail. So showing a client the alternative before they decide costs you review time, not a second export.

So the clause is not protecting your credit spend on variants. It is protecting the direction, which is the scarce input. That distinction is worth being explicit about, because "each version costs me money" invites the client to offer to pay for versions, and that is not the deal you want. The deal you want is unlimited versions against a locked direction, and a real price on changing the direction.

Two habits make it work. Run review on share links rather than file handoffs so iterating genuinely feels fast — a client review loop built on previews and share links is the mechanics, and sharing a generation with a link is the primitive. And log which trigger fired on each job: a shop where trigger (d) fires constantly has an intake problem, not a contract problem.

FAQ

Is it fair to charge for a change when producing the change is nearly free?

You are not charging for production, and the clause should not imply you are. You are charging for the work the change invalidates and the direction work it requires — it costs something because approved work is being thrown away, not because a render is being run. Stating it that way tends to end the objection, because the client can verify it against the ladder they signed.

How does this interact with a rounds cap?

They cover different things and you want both. Rounds govern iteration inside a locked direction — how many passes of notes you absorb before it becomes a scheduling problem. Change orders govern direction itself. A revision policy that stops scope creep covers the rounds half, including where the line between a revision and a new deliverable sits.

Should the volume rate for extra outputs be zero?

For small numbers against a locked direction, often yes, and saying so up front buys goodwill. Set it above zero once the volume implies real review time, because reviewing forty variants is not free even when generating them nearly is. That is where the cost has moved: filtering, governance and taste, not production.