How do you rebuild a creative workflow around a technology that's only two weeks old?

01    This pilot came with a hard deadline, therefore, needed a faster rollout.

The deeper problem wasn't really about AI at all: the existing workflow put the burden of coordinating on the director, with no structure for when input was locked in or how work moved between stages. Any process needed to fix that first, and only then could AI be added inside it.

The as-is workflow.

02    Offloading the strategizing to the studio and protecting employee mental health and creativity.

A core issue raised by studio employees and nearly every designer interviewed: they wanted to be treated as professionals, not micromanaged. In practice, that means if you're contracting a studio, give them the brief and let them bring you something concrete before you start judging it.

In reality, no one waits; They pile on pressure. And When AI approaches you like a boogeyman, the pressure compounds and burnout follows.

The toolset was simple — Google Docs for the brief, Google Sheets for shot tracking, Frame.io for review, and Adobe Photoshop's Generative Fill for design — using only what the studio already had, so the process could be validated before justifying any further tool investment.

03    The workflow ran in five stages.

It opened with a facilitated tone-alignment session to internalize the director's vision — the highest-risk point in the process.

From there, the project manager and design lead broke the project down shot by shot, deciding on what part of the shot AI was needed.

The to-be workflow.

Generative Fill had launched in beta only two weeks before the pilot began, so the design lead had to build evaluation criteria for it in real time.

Meanwhile, junior designers used it to mass-produce environments and backgrounds while the sketch artists handled character work by hand, deliberately keeping the juniors focused on building design judgment rather than dependency on a tool.

Every AI-generated asset then went through rigorous compositing and QA — intial outputs were usable as a starting point but still required manual rework, and designers had to assess each asset before it moved forward, something the studio had communicated to the client early on.

04    The threshold for using AI was simple.

If compositing an AI output took as long as producing it manually, it wasn't worth using AI at all. But, the value of AI wasn't that it eliminated skilled work — every generated asset still needed rework before it was usable.

The core benefit we saw was that designers no longer lost time searching for the right reference or asset, and could get to a usable starting point faster.

Approved assets were exported as layered passes to the edit team, who assembled the final cut for director review and sign-off — this was the studio's actual client-facing deliverable.