A Client Approval Workflow for AI Creative

The fastest way to lose money on AI creative is to show a client a finished video too early. They react to the wrong layer (a color, a background detail) when the real decision was the concept two stages back, you re-render the whole thing, and the round count balloons. AI generation made producing options nearly free, which quietly made this worse: when a variation costs a dollar, it's tempting to skip the cheap approval stages and just render finals. The fix is a staged approval workflow with clear gates, so the client signs off on the cheap-to-change decisions first and the expensive-to-change ones last.

This guide lays out that workflow: four approval stages from moodboard to final, what to show the client at each one, why showing the right thing at the right time controls scope creep, and an approval checklist that includes the AI disclosure conversation. It's written for agencies and freelancers delivering AI creative to clients, but any team with an approver and a maker can run it.

Why staged approval beats "show them the video"

Every creative decision has a cost of change. Concept is cheap to change and gets more expensive the further you go: reworking a concept after moodboard is free; reworking it after the final is rendered means throwing away everything downstream. The point of staged approval is to get sign-off on each decision while it's still cheap to change.

The trap AI creates is that finals are so cheap to produce that skipping stages feels efficient. It isn't. If you show a polished animatic before the client has approved the concept, and they don't like the concept, you've spent generation credits and, worse, anchored them on a direction you now have to unwind. Feedback also expands to fill whatever fidelity you present: show a rough frame and you get concept notes; show a finished clip and you get notes on grain and kerning. Staged approval matches the fidelity of what you show to the decision you actually need, which is what keeps rounds bounded.

Run the whole thing on a shared creative canvas so every stage, every version, and every client comment lives on one surface. That alone removes most version chaos, because there's no exported final_v4 floating in an email.

The four approval stages

Stage 1: Moodboard (approve the direction)

What you show: references, a color and lighting direction, tone, a written line on the concept. Pinned stills, not motion. Cost: image generations only, pennies.

What you're getting approved: the creative direction, before you've generated anything expensive. This is the cheapest gate and the most important. If the client wanted "warm and premium" and your board reads "clinical and cool," you find out now, at the cost of a few reference images, not after a hero render.

Get explicit sign-off here: "Does this direction feel right before we start producing?" A yes at Stage 1 is what lets you push back later when a client tries to relitigate the whole look at the final stage.

Stage 2: Stills (approve the look)

What you show: key frames rendered in the approved direction. The hero shot, the product framing, the composition, character look if there's talent. Still images, generated with something fast and cheap like Nano Banana Pro at $0.04-0.08 each.

What you're getting approved: the actual visual execution, framing, composition, and product accuracy, while it's still a still image and therefore cheap to regenerate. Nail character consistency and product likeness here, because fixing them after animation costs far more. A client comment like "the label is wrong" should surface at Stage 2, never at final.

Stage 3: Animatic (approve the motion and timing)

What you show: a rough motion pass. The approved stills animated, the sequence, the pacing, where the cuts land. This is an animatic: motion and timing, not final polish. Use a fast, affordable model like Kling 3.0 at $0.28-0.40 per clip.

What you're getting approved: does the thing move and cut the way it should. Structural motion notes ("hold the product shot longer," "cut to the CTA sooner") belong here, where re-rolling a clip is cheap. You're deliberately not asking for final-quality judgment yet; if the client fixates on polish, redirect them: "this is a timing pass, we'll finish the look once the sequence is locked."

Stage 4: Final (approve for delivery)

What you show: the finished, upscaled, graded deliverable in the target format and aspect ratio. Full quality. This is the only stage where you spend the premium compute (a Veo 3.1 hero at $0.85-1.20, an upscale on the winners).

What you're getting approved: delivery. Because the concept, look, and timing are already locked from Stages 1-3, the final review should be a formality plus final polish notes, not a re-open of settled decisions. If someone tries to reopen the concept here, you have three prior sign-offs to point to. That's the scope-creep firewall the staged workflow buys you.

How the stages control scope creep

Scope creep in creative is usually a decision reopening after it was made. Staged approval prevents it structurally: each gate has an explicit sign-off, and each sign-off is leverage. When a client at Stage 4 says "actually, can we make it feel more energetic," you can say "we locked the direction at moodboard and the pacing at the animatic; an energy change means reopening those stages, which is a new round." That's a fair, documented conversation, not a fight. Without stages, every note feels equally valid at every moment, and the round count is unbounded.

The other half is fidelity matching. By showing exactly enough polish to make each decision and no more, you stop inviting feedback about layers that aren't up for decision yet. That is the single biggest lever on round count, and it's free.

The approval checklist

Before you send any stage to a client, run this:

• Right fidelity for the decision. Moodboard = references, Stills = key frames, Animatic = rough motion, Final = finished. Never show more polish than the current decision needs.
• Explicit sign-off requested. Each stage ends with a clear yes/no question, not "let me know your thoughts." Ambiguous asks produce ambiguous approvals.
• Prior approvals referenced. Remind the client what's already locked, so the review stays scoped to the current gate.
• Product and brand accuracy verified. Especially at Stills and Final: correct product, logo, colors, and no warped details. Review on the canvas at full resolution, not on a thumbnail.
• AI disclosure handled. Confirm with the client how the creative is labeled. Whether and how AI-generated ads must be disclosed varies by platform and region; agree on the approach explicitly rather than assuming. See do you have to disclose AI-generated ads for the current landscape.
• Deliverable specs confirmed. Aspect ratio, duration, and platform format match the brief before final render, not after.

Worked example: a 30-second product ad

An agency delivering a hero ad for a DTC skincare brand:

1. Moodboard (day 1): a board of warm, sunlit bathroom references, product pinned, one-line concept. Client approves the direction. Cost: a few image gens.
2. Stills (day 2): three key frames, product hero and two lifestyle beats, in Nano Banana Pro. Client flags the jar cap color; the maker re-rolls that frame. Approved. Cost: under $1.
3. Animatic (day 3): the frames animated in Kling 3.0 into a rough 30-second cut. Client asks to hold the product shot two beats longer. Re-cut, approved. Cost: a few dollars of clips.
4. Final (day 4): the hero beat re-rendered in Veo 3.1, everything upscaled and graded, delivered 9:16 and 1:1. Client signs off with two small polish notes. Cost: premium compute only on the locked-in cut.

Four rounds, four gates, no reopened decisions, and the expensive compute spent once at the end. Compare that to showing a finished cut on day 2 and discovering the client hated the concept.

Related reading

To structure the multiplayer side of this, see the collaborative AI canvas for teams. For the conversation that comes before the work, how to pitch an AI campaign to clients covers setting expectations up front, which makes every approval stage smoother.

---

Run every approval stage on one surface, from moodboard to final. Explore the 8frame canvas and workflow library to build the staged workflow your clients sign off on.