How many options to show a client
Showing everything you generated hands curation to the client. A rule for how many directions to present at each gate, and what to do with the rest.
You generated forty. You show three. The gap between those numbers is not waste — it is the job.
The temptation to close it runs the other way, because showing everything feels generous and transparent, and because the pile exists and deleting it feels like throwing away money already spent. It is neither generous nor transparent. It moves the hardest part of the work — deciding which of these is right — onto the person who hired you specifically so they would not have to do it, and it does so without charging for the transfer.
What a big pile of options actually does to a review
It slows the decision and lowers its quality. A reviewer with three options compares them. A reviewer with twelve skims, forms an impression of the set, and picks the one that stands out — which is a measure of contrast against the other eleven, not of fit to the brief.
It invites assembly. The predictable outcome of showing many finished-looking versions is "can we take the opening of the first one and the ending of the fourth." That request sounds cheap and is not: on a generated piece it usually means new shots, new pacing, and a new grade, because the two versions were never built to interlock.
It costs you the part of the role that is still scarce. The consistent description of how senior creative work changed is a shift from creator to curator: designing the inputs, filtering the outputs, enforcing brand voice across them. Handing over an unsorted pile is opting out of exactly that. It also reads as uncertainty, because it is.
Every option you show stays live forever. Something rejected in round one comes back in round four, usually from a stakeholder who was not there. Work the client never saw cannot be resurrected by anyone.
None of this is an argument against generating widely. Generate as widely as the budget allows — the Nike work with Serena Williams reportedly generated 130,000 virtual tennis matches, with AI used for exploration and human craft applied to the finish. That is the correct shape: enormous exploration, tiny presentation.
The rule, gate by gate
| Gate | Present | Format | What you say |
|---|---|---|---|
| Concept | 2 directions | One page each: name, one-line idea, reference board | "Two directions. I'd run A." |
| Key frame | 1 per shot | The still, in shot order | "This is the look. One alternate on the hero shot." |
| Motion | 1 per shot | Raw clips, no music or captions | "Approving the performance and camera here." |
| Export | 1 cut | 480p preview | "This is the cut. Notes on timing and text." |
Two at the concept gate, not one and not four. One direction reads as take-it-or-leave-it and pushes the client into inventing a third themselves, which is the most expensive kind of third. Four invites averaging: the client likes something in each, and you end up building a committee edit that no single person would have chosen. Two forces a real comparison, keeps you visibly in the position of having chosen, and gives the conversation somewhere to go if the first answer is no.
Make the two genuinely different. Two variations on the same idea is one option presented twice, and clients can tell.
One per shot at the key-frame and motion gates. The exception is the shot carrying the risk — the face, the hero product beat, the thing you privately are not sure about. Give that one an alternate and say why: "Two versions of the opening because it's the shot doing the most work." A named exception reinforces that the single options elsewhere are decisions, not defaults. Storyboarding with AI images first is where these stills come from, and each one that gets approved is a shot you are not going to animate twice.
One cut at the export gate, always. Two finished cuts is the single most expensive presentation you can make, because it reopens the assembly after everything upstream was signed off.
Platform variants are not options
A 9:16, a 1:1 and a 16:9 of the same idea are three deliverables, not three choices. So are a 15-second and a 30-second edit, and so are language versions. Presenting them as options invites the client to pick one and quietly forget they were paying for three; presenting them as deliverables keeps them on the invoice where they belong.
Present with a recommendation attached
The format that works is short and takes a position:
"Two directions attached.
A — 'Quiet proof.' One product, one claim, no voiceover. Slower, and it looks more expensive than it was. I'd run this one; it's closest to the reference board you approved and it works with sound off.
B — 'Fast facts.' Three benefits in fifteen seconds, captions carrying it. Louder, better for a cold audience, further from your usual tone.
Happy to go either way — if you don't have a strong view, I'll build A."
Three things in there earn their place. A name per direction, so later conversations have a handle. A stated recommendation, because withholding one is where authority leaks. And a default if nobody answers, which is what keeps the gate from becoming a stall. Share links are the delivery mechanism — they open with a preview card and no login, which is most of what "easy to review" means when the reviewer is on a phone.
What to do with the thirty-seven you did not show
Keep a shortlist, not the pile. Save the five or six near-misses into a named project so they are findable. Saving generations to a project does not spend generation credits and skips anything already saved, so a shortlist is cheap to maintain. When round three asks "is there a version with more of the product," you either have it or you know quickly that you do not.
Promote the reusable ones into a swipe file. A rejected direction for one client is often the right answer for the next brief. Building a video swipe file covers keeping that organised enough to actually search.
Delete the rest. Storage is not the problem; findability is. A library where everything was kept is a library where nothing gets found.
Treat unshown directions as inventory. If a client asks for something new mid-project and you already have a strong unshown option, that is not a free consolation prize — it is a direction change, and it should be quoted as one.
Generate widely, internally
The volume belongs on your side of the wall. The agent can fan a single prompt across several named models in one request, which turns "which model handles this shot best" into one comparison instead of four sequential experiments. A/B testing AI models with one prompt covers running it, and a scoring rubric for comparing video models covers judging the results with something more repeatable than a preference.
The one case where volume should be client-facing is paid testing, because there the buyer is a media budget rather than a taste. Even then the client approves the matrix — how many hooks, how many variants, what is being tested — not each individual asset. A/B testing video creative properly and the ad creative iteration loop cover designing that so the results mean something.
FAQ
What if the client explicitly asks to see everything?
Ask what they are looking for, because the request usually means something specific: they are nervous about a particular element, or a stakeholder wants to feel involved. Answer that instead. "I'll send the full set of hero-shot options" is a useful concession; "here is the folder" is not.
Does presenting two directions mean I did twice the work for free?
No, and pricing should make that visible. Two directions at the concept gate are two pages and two reference boards, not two productions. Only one gets built. If the client wants both built, that is a second deliverable.
What if they like neither of the two?
Then the concept gate did its job cheaply, which is the entire reason it exists. Go back to the reference board with the specific reason both failed and bring two more. Nothing has been generated, animated or graded yet.
Should I show my process — the rejected versions — to demonstrate effort?
Show the volume as a number if it helps ("we explored about forty framings for this shot"), never as a folder. The claim is credible without being browsable, and browsable is what reopens decisions you already closed.