On our homepage we describe the process in three steps, because three is what fits on a homepage. Clients who are about to hand over a budget usually want the longer version, so here it is in full.
This is the actual pipeline. Not a diagram made for a pitch deck, but the sequence a project moves through here in Dubai, in order, with the points where a client is asked to look at something and say yes.
One thing worth noting before the list. Generation, the part everyone thinks of as the whole job, is step five of seven. By the time we ask a model for a moving frame, almost every decision that determines how the film looks has already been made and signed off. That ordering is not a stylistic preference. It is the difference between a film and a folder of clips.
STEP 1: INTERROGATE THE BRIEF
Most briefs arrive describing a deliverable. Sixty seconds, four scenes, needs to feel premium, needed by the fifteenth. That tells us what to build and nothing about whether building it is the right idea.
So the first conversation is about the job the film has to do. Who is watching, in what context, and what do they have to feel or believe by the end. Is this the first time the audience meets the brand or the fifth. Is it running before a purchase decision or after one. Does it need to survive being watched with the sound off.
This is also where we test the method. We run live crews as well as AI, and a meaningful share of incoming briefs are better served by a camera. If yours is one of them we say so at step one rather than at step six. We wrote out that whole decision in AI vs traditional production.
What comes out of step one is a one paragraph statement of intent that everything downstream gets measured against. It sounds like an unnecessary formality until a shot is up for debate in week two and there is an agreed sentence to settle it with.
STEP 2: TREATMENT AND SHOT LIST
Now a director writes the film. Structure, beat by beat. What the opening frame has to establish. Where the turn happens. What the last three seconds leave behind.
Out of that comes a shot list, and the shot list is the document the entire project runs on. Every shot gets a number, a purpose, a rough duration, a lens intention, a camera move if it needs one, and a note on what has to be consistent with the shots either side of it.
The shot list is also the pricing document. Shot count is the main thing that moves a quote, which we set out in detail in what AI video actually costs in Dubai. Once the shot list is agreed, the number is agreed, and neither drifts quietly afterwards.
Client approval gate one. You approve the treatment and the shot list. Changing the plan here costs a conversation. Changing it after step five costs money.
"By the time a model produces its first moving frame, the film has already been directed. Generation executes a decision, it does not make one."
STEP 3: ART DIRECTION AND LOOK DEVELOPMENT
Before anything is generated, an art director builds the visual rules the film will obey. Palette, and specifically which two or three colours are allowed to carry weight. Wardrobe and materials. Time of day and the direction light comes from. Skin tone targets. Texture, grain, how clean or how worn the world is.
In practice this is a reference board plus a written set of constraints, and the written part matters more than people expect. A board is open to interpretation. A line like "single hard key from camera left, warm practicals, no fill" is not.
These rules then get carried into every prompt, every reference image and every correction for the rest of the project. This is the step that separates a film with a look from a film where each shot has its own opinion.
STEP 4: FRAME APPROVAL, STILLS BEFORE MOTION
This is the step most studios skip, and skipping it is visible in the result.
Every shot on the list gets built as a still first. Composition, framing, subject, wardrobe, light, background, lens character. We iterate on the still until it is genuinely the frame we want, then we do it again for the next shot, checking each one against the ones around it for continuity of face, palette and light.
The reason is economic as much as creative. A still takes seconds to regenerate and costs almost nothing. A moving shot takes far longer, costs more, and if the composition was wrong the motion was wasted. Fixing a problem at still stage is roughly the cheapest correction available anywhere in the pipeline.
It is also where character consistency is won or lost. If the same person appears in eight scenes, the eight approved stills are what hold that person together. Trying to enforce consistency later, across moving footage, is a much harder problem with much worse results.
Client approval gate two. You see the approved frames. What you are looking at is close to the finished film in still form, which means there are very few surprises left after this point.
STEP 5: GENERATION
Only now does anything move. Each approved still becomes the anchor for its shot, and the model is asked for motion within the boundaries already set: this frame, this move, this duration, this behaviour.
Constrained work like this succeeds far more often than open generation, because the model is solving a narrow problem rather than inventing a scene. Even so, not every attempt is usable. Shots get run multiple times and the takes get judged the way takes have always been judged, on whether the motion is believable and whether it serves the cut.
We use different platforms for different jobs, because they genuinely have different strengths, and part of the work at this stage is knowing which tool suits which shot. That choice sits with us, not with the client, in the same way a DOP picks the lens.
Anything that fails at this stage goes back to step four rather than getting patched. Patching a broken shot is how a film starts to look assembled.
STEP 6: EDIT AND SOUND
The clips become a film here, and this is where most work made with these tools falls down. Individually decent shots, cut with no sense of rhythm, produce something that feels flat even when nothing in it is technically wrong.
So an editor cuts for pace: how long a frame holds before the eye has taken it, where a cut lands against the track, when to sit still and when to move. Shots get trimmed, reordered, and sometimes dropped entirely even after being generated, because the film is better without them.
Sound is treated as part of the picture, not a layer added at the end. Music selection or composition, room tone, foley detail, and where there is dialogue, voice casting and sync. Sound is the fastest way to make a sequence feel physically real, and its absence is the fastest way to make it feel like a demo.
Client approval gate three. You see the first cut. Two revision rounds are included in every project, and they belong here, where changes are cheap and meaningful.
STEP 7: GRADE, FINISH AND DELIVERY
The whole film goes through a colourist. Not a filter, a grade: matching shots to each other, shaping contrast, holding skin tones, deciding where the eye is pulled in each frame.
This step does more to unify a film than anything except the frame approvals in step four. Shots made at different times, on different platforms, with different quirks, are brought under one consistent hand. It is also where the last traces of the process get cleaned up, which is the subject of what makes AI video look like AI.
Then delivery, in every format the media plan needs. Sixteen by nine, nine by sixteen, one by one, four by five. Cutdowns. Language versions. Platform specific exports with the right codec and the right loudness.
Variants are worth planning for at step two rather than requesting at step seven. The creative cost sits in the first version, so the eighth placement is a fraction of the first. Ask for them late and you pay for a second pass through the pipeline instead.
WHY SEVEN STEPS AND NOT TWO
You could compress this. Take the brief, generate, send. Some studios in this market do exactly that, and their pricing reflects it honestly enough.
The problem is that a two step process has no place to make a decision in. Direction happens at step two. Coherence happens at step three. Cost control and quality control both happen at step four. Feeling happens at step six. Unity happens at step seven. Remove those and what is left is generation, which is the one part of this that a model does on its own.
Seventeen years of production taught us that most of filmmaking is decided before anyone rolls, and cleaned up after everyone wraps. The tools changed. That part did not.
