For most of my working life, usable video started with someone in a room holding a camera: choosing the angle, lighting the space, waiting for the right moment. I've spent years doing that work, photographing properties and interiors, filming corporate events, then turning the raw material into something worth watching in Adobe Premiere Pro.
Recently I've started bringing AI-generated footage into that workflow. By AI-enhanced video production, I mean using AI to generate some or all of the footage, sometimes including an AI-generated presenter, while the brief, script, review and edit are still directed and finished by a person. This article grew out of a recent AI-assisted commercial video workflow, combining AI-generated scenes with traditional editing. It's about the process and what I learned, whether the end result is a promotional video, a product presentation or a UGC-style social ad.
What struck me most is how much of the job stayed the same. Generating a clip is now fast. Producing a convincing finished video still takes a clear brief, a plan, careful review and a proper edit. AI has changed one part of production. It hasn't removed the rest.
The creative brief still comes first
It's tempting to start with the tool. In my experience the better first question is the one I'd ask before any shoot: who is this for, and what should they think, feel or do once they've watched it?
That's where the work started: the audience, what the video needed to achieve, concepts and the message. Then the structure: an opening hook strong enough to stop someone scrolling, a short middle that delivers one idea, and a closing call to action. I studied how comparable short-form adverts are built and weighed visual styles, presenter approaches and ways of showing the product. Only then did I write the script and scene-by-scene directions.
None of this is new; it's what pre-production has always been. If anything, it matters more with AI. A generation tool can't guess what you meant. If you don't know what each scene is for, you get footage that looks impressive and says nothing in particular.
Generating footage is not producing a video
A striking AI-generated clip is easy to admire and easy to share. That doesn't make it an effective advertisement.
An advert has a job: hold attention, get one idea across, look as though it belongs to the brand, and leave the viewer knowing what to do next. A beautiful clip that drifts off-message, or changes the product's shape halfway through, works against that job.
Continuity is where I spent most of my review time, checking every scene for generation errors, unnatural movement and continuity problems. Details a film crew keeps fixed without thinking, like lighting, wardrobe, background and an object's proportions, can shift between generations, and within a shot, hands and motion can look subtly wrong. Researchers see the same pattern: benchmark studies such as VBench-2.0 (opens in a new tab) and T2VPhysBench (opens in a new tab) find that current video models still struggle with realistic anatomy, physics and commonsense movement, even when the result looks plausible at a glance.
AI-generated presenters with spoken dialogue add another layer. When someone talks to camera, viewers watch their face closely, so small mismatches between lips and words, or delivery that feels flat, are noticed quickly. Those were the scenes I looked at hardest.
Why I build videos scene by scene
Rather than asking a tool for one long video, I generate each scene separately and assemble them in the edit.
Each generation then has a single job, so it's easier to describe and to judge. When one scene fails, I regenerate that scene alone and keep the parts that were right. Because generation is paid for in credits on the services I've used, redoing a short shot costs less than redoing a sequence. The same visual direction and references can be applied to every scene. And separate clips give the edit room to breathe: I can reorder, trim or hold a moment longer without going back to the generator.
It isn't a universal rule. If an idea depends on one continuous movement, such as an unbroken walk through a property, cutting it into pieces can break what makes it work, and joining generated shots seamlessly is hard. More scenes also mean more continuity to manage. Modular production suits work built from distinct beats, like a hook, a demonstration, a message and a call to action. That describes a lot of short-form advertising, but not all of it.
Reference images, consistency and control
Reference images are one of the most useful ways to steer generated footage. Instead of describing everything in words, you show the tool what the product, space, lighting or presenter should look like. In this workflow I prepared reference images to guide the visuals and keep scenes consistent.
This is where photography experience earns its keep. A good reference image is a photographic decision: composition, the direction and softness of light, colour temperature, what's in the frame and what's left out. More than 2,750 professional photography assignments, most of them properties and interiors, taught me how much a room changes with a small shift in angle or light. That judgement now goes into the images I give a generator.
But a reference guides the result; it doesn't guarantee it. Generated footage can still alter a label, distort a logo or change a small detail that matters to the brand. Every scene still has to be checked against the brief. A good reference raises the starting point. It doesn't replace the review.
The video is made in the edit
However good the generated scenes are, they arrive as separate pieces. They become a video in the edit, which for me still means Premiere Pro.
The first decisions are about selection: not every scene that technically works deserves a place. Then come sequencing and timing, which in short-form work are measured in frames. A hook that runs half a second too long loses people; a call to action that arrives too early feels pushy. Most cuts should be invisible, because an elaborate transition usually draws attention to the join. Branding and the closing frame need to feel part of the film, not stuck on the end.
In this workflow, that assembly is what I've completed so far: selecting and sequencing scenes, adjusting timing and pacing, integrating the branding and closing elements, and combining everything into one coherent piece. A finished commercial needs more, and these are the finishing stages I plan for.
Audio. A voice that shifts slightly in tone between separately generated scenes is jarring, so voiceover continuity has to be checked across the whole cut. Music and sound design do much of the work of setting pace and mood.
Colour. Separately generated clips rarely match exactly. Colour correction makes a set of scenes look like one film.
Captions. They carry the message for anyone watching with the sound off, and need to be timed and placed so they don't fight the picture.
Delivery. Vertical, square and widescreen versions each need their own framing. Reframing for a different shape is a composition decision, not a resize. Premiere Pro can also attach Content Credentials (opens in a new tab) at export, a provenance record of how the file was made, although Adobe notes that not every platform keeps them.
None of this is specific to AI. It's ordinary post-production, and it's where separate clips become a production.
Where UGC-style advertising fits in
UGC stands for user-generated content: videos, photos and reviews made by real customers, users or independent creators rather than by a brand. Much of its appeal comes from that origin. Someone talking plainly to a phone camera about something they've used often feels more believable than a polished advert.
UGC-style advertising borrows that approachable, conversational presentation: a person speaking directly to camera, vertical framing, a relaxed delivery. The difference is that it's produced as advertising. Brands have long worked with creators to make content like this, and AI-generated presenters now make it possible to produce some UGC-style formats without filming anyone. That is usually what people mean by AI UGC or AI UGC video ads.
The distinction matters. Genuine user-generated content reflects a real person's real experience. An AI-generated presenter has no experience of the product, so an AI-generated UGC-style ad shouldn't be presented as a customer testimonial or as a real customer's story. In the US, for example, the FTC's rule on consumer reviews and testimonials (opens in a new tab) prohibits testimonials that misrepresent who is giving them or whether that person has actually used the product, including AI-generated ones. Where it isn't obvious that a presenter is AI-generated, saying so can also help avoid a misleading impression.
Used honestly, the format suits short-form promotional videos and social advertising: explaining a product, demonstrating a feature or introducing an offer in a voice that feels more personal than a narrator. The craft is the same as everything above. The script has to sound like a person talking, not a brochure being read aloud. The presenter has to stay consistent from scene to scene, the delivery has to be credible, the branding has to fit, and the edit has to keep the pace short-form viewers expect. A presenter who looks almost right but sounds scripted undermines the informality the format depends on.
What AI changes for smaller businesses
For a business that needs promotional video but can't justify a shoot with crew, location, presenter and travel, AI-assisted production opens real possibilities. Ideas can be visualised before anyone commits to them, some physical requirements fall away, revisions are more flexible, and the work can run remotely. Visual experiments that would never have been filmed can become affordable, from short-form social ads to UGC-style product videos.
The limitations are just as real. Generation costs add up when scenes need several attempts, and review and editing take time; sometimes a half-day shoot would have been faster and cheaper. Accuracy matters too. The UK's advertising regulator, for example, says its rules apply however an ad is made, and that disclosing AI use is very unlikely to fix an ad that is fundamentally misleading, such as an AI-generated image that exaggerates what a product does (ASA, 2025 (opens in a new tab)).
Rights need attention as well. Usage terms vary between services, something I've also written about from a photographer's side in what happens to a photograph after you share it. Ownership is less settled than many assume: the US Copyright Office's position is that prompts alone generally don't make someone the author of the output, although human selection, arrangement and modification can be protected (US Copyright Office, 2025 (opens in a new tab)). Realistic people and voices raise questions of likeness and consent.
Then there's disclosure. YouTube (opens in a new tab) and TikTok (opens in a new tab) require creators to label realistic AI-generated content, and Meta (opens in a new tab) requires disclosure in certain political and social-issue ads. In the EU, the AI Act has required clear labelling of deepfakes since 2 August 2026 (European Commission (opens in a new tab)). The rules depend on where and how a video runs, so they belong in the plan from the start, not in a check at export.
And some subjects should simply be filmed. If a business sells something real, like a holiday home, a hotel room, an event or its own team, viewers deserve to see the real thing. Much of my photography career has been about showing spaces honestly, and I don't expect AI to change that.
The creative professional's role is changing
What I've noticed most is a shift in where the skill sits. Less of my time goes into operating equipment and more into directing: writing scene directions, preparing references, judging footage and making the edit work. Those tasks draw on what I've always relied on: knowing what a good shot looks like, why one cut works and another doesn't, and when something on screen is slightly wrong before you can name why.
I've written more broadly about how AI changes what experience can deliver; video is where I see it most clearly. The tools will keep changing, probably faster than this article ages. The questions haven't. Who is the video for? What should they take from it? Does every second earn its place?
If you're considering a short promotional video or a UGC-style ad and aren't sure whether it should be filmed, generated or a mix of both, I'm happy to talk it through. Get in touch.