The 7-Stage
AI-VFX Pipeline

There is a version of AI filmmaking that lives entirely inside a text box. You type, you wait, an image arrives, and for a moment it looks like the whole problem has been solved. I understand the appeal, because I felt it too. But a commercial is not an image. It is a sequence of decisions that have to survive a client, a broadcaster, a brand guardian, and a delivery spec. The generation itself turns out to be one stage among seven. What follows is the order I actually work in. The order matters: my generations stay controllable because I build the control before I generate, not after.

01 · Concept & DirectionDeciding what the film is before a single frame exists

Everything downstream inherits its constraints from this stage, so it is worth being slow here. Concept work fixes the non-negotiables: what the brand is actually selling, what must remain literally true about the product, the emotional register, and which shots carry the argument. On the Gigaclear commercial the immovable facts were set first. Real actors would live inside these environments. A crowd of roughly a hundred had to read as a coherent community. And the whole thing had to hold on national broadcast. Those decisions were made in language and reference before any model was touched. When the tool sets the direction, a striking test frame can end up dictating the concept. It looks efficient at first, but it usually creates expensive corrections later.

02 · PrevisAI's real superpower, and its honest limit

This is where AI earns its place first. Generative models let me build a previs, a full visual treatment of the film. It can arrive in hours rather than weeks, letting a client see the mood, light, framing and overall feel of the spot before anyone commits the larger budget.

But previs is exactly that: a preview. It is high quality and it is not production-ready, and being honest about that distinction is half the job. The previs proves the idea and aligns everyone on it. It does not yet have the consistency, the control, or the exactness a broadcast deliverable needs. So the previs is not the answer. It is the brief for the stages that follow, the thing the rest of the pipeline is built to make real and repeatable.

03 · 3DBuilding the control before the generation

This is the stage that separates a controllable pipeline from a hopeful one, and it comes before generation on purpose. Once the previs is approved, I build the parts that have to stay exactly themselves as real 3D. That work happens in Cinema 4D and the wider 3D toolset. For Gigaclear, the hundred-character crowd was rebuilt as actual geometry. Each character could be posed, lit, and moved as a controlled scene, not re-rolled and hoped over. Products, key environments, and anything that needs a precise, repeatable camera move are built or laid out here too.

The point is control. A 3D scene is something you can direct, the same camera, the same layout, the same identity from every angle. It becomes the skeleton the generation is wrapped around, which is how you get the beauty of AI without surrendering to its habit of reinventing everything each frame.

04 · GenerationGenerating on top of the 3D

Now the model does what it is good at, but against a clear target. Rather than generating a shot from a sentence and hoping it matches the last one, I generate on top of the 3D: using the rendered layout, depth, and geometry to condition the generation so the AI adds richness, texture, light and life while the underlying structure stays locked. The 3D holds the composition and the identity; the model supplies the photographic treatment. Used this way, generation can add atmosphere and detail without the drift that makes prompt-only work unreliable when an element has to remain the same twice.

05 · Visual EffectsCleanup, stitching, and the major fixes

No generation arrives finished. In this pipeline, VFX is the stage that turns approved material into shots that can survive an edit and delivery. The detailed work—temporal cleanup, compositing, tracking and integration—has its own explanation in Why AI Video Still Needs Traditional VFX.

06 · Grade & FinishingMaking seven stages read as one film

Once the frames hold, the film has to feel like one film. A proper grade pulls the generated, rendered, and shot material into a single palette and mood, correcting the tonal drift that can make a warm shot and a cold shot feel as if they belong to different afternoons. Sound design and the final edit carry as much of the emotion as the picture. Without finishing, the sequence can be technically clean but tonally inconsistent, with each shot lit by a slightly different sun. The grade is not decoration. It is what makes the whole thing read as a single intention.

07 · DeliveryThe unglamorous stage that decides whether it airs

A film that cannot be delivered to spec does not exist as far as a broadcaster is concerned. This stage is versions and compliance: correct resolution and frame rate, legal audio loudness, safe areas, colour space and legal levels, clearance-ready masters, plus the fleet of cutdowns and aspect ratios a modern campaign consumes.

• Broadcast master to the network's exact technical spec, with correct loudness and legal levels.
• Social and online versions in their own ratios and durations, reframed rather than crudely cropped.
• Textless and versioned masters so the spot can be localised or amended without a rebuild.

A pipeline that treats generation as the finish line can arrive here with one beautiful file and no way to conform it. Delivery tests whether the work can be used, and it only gets here intact because the control was built in stage three, before the model was asked to generate a frame.