02. The Challenge
Real people are the strictest test there is
Three real people have to look like themselves. With invented characters, nobody notices if a face comes out slightly differently in one scene. Here, the people in the film were sitting at the client’s table: the two staff from Kiel who deliver the machines, and the owner of Bikerbar Silberwolf who receives them at the end. The feedback on the first version was as short as it was unambiguous: “That doesn’t look like me.” A member of staff answering the phone at the counter has to be recognised instantly by his colleagues. Two staff rolling a machine off the van have to be the same two at every delivery. A bar owner shaking one of their hands in the courtyard has to be unmistakable to his own guests. Generative models happily produce a friendly bearded man in his late twenties. We needed precisely this one.
The van is a real vehicle with real lettering. White high-roof panel van, narrow grey sill band, the bar-and-shield as an outline on the flank, HD KIEL in black capitals. That livery has to be legible and correct in every shot: on the forecourt, at the harbour, outside the chalet, on the autobahn. AI models alter lettering with every generation. For a brand, every deviation is an error.
The dealership exists and can be checked. The counter with its stone front and the large bar-and-shield, the monitors, the ceiling tiles carrying the emblem, the forecourt with the “Harley-Davidson Vertretung Kiel” gazebo, the rows of bikes on display. Anyone who knows the building sees every deviation.
Every motorcycle is a specific model. The matte black Fat Boy, the Heritage Softail with windshield and saddlebags, the orange custom, the Sport Glide going back into the van as a trade-in. Not “a Harley”, but this one each time.
Hands on machines. Two men rolling a motorcycle down a ramp out of a van is a stress test for generative video. Who holds the bars, who supports the rear, which way does the wheel roll, who is on the ramp and who is beside it. This kind of interaction between person and object is where the models fail most often: a third arm, a rider suddenly sitting on the machine, a wheel that slides rather than turns.
Five locations, one van. The film travels from Kiel across the whole of Germany: the warehouse district at Hamburg harbour at dusk, a cobbled alley, a mountain chalet in evening light, the autobahn, the Silberwolf courtyard. The same van, the same two staff, at every location.
A scene from the AI commercial.
03. The Approach
First we documented the dealership
Before an image was generated we built Harley-Davidson Kiel as a reference library: the counter from several sides, the showroom floor, the forecourt, the van from the front, from the side and with its rear doors open. Several portraits were available for every member of staff appearing in the film. Each motorcycle was referenced from its own photograph. The bar-and-shield was available as a clean file.
Those references were the basis for every generation that followed. We didn’t describe the dealership to the models. We showed them.
Dramaturgy: from the call to the arrival
The film follows a simple line. “Some journeys begin with a voice in your head, others with a phone call, and some with just a click.” At the counter, a member of staff picks up the phone. The van leaves the forecourt. Then three deliveries in three very different places: the Heritage at Hamburg harbour in the blue hour, the Fat Boy in an old alley, the orange custom outside a mountain chalet in the evening sun. At each delivery the same two staff roll the machine down the ramp while the customer stands beside them. “And the old machine? We take it with us.” Back in Kiel, the Sport Glide is rolled off the van.
Then the film changes perspective. “From here on, you don’t ride alone.” The three customers are now on their machines, riding with other riders on the autobahn toward Brandenburg. “Many roads. One destination.” They roll into the courtyard of Bikerbar Silberwolf, where the bar’s owner greets one of the two Kiel staff with a handshake: two real people who genuinely know each other, in a picture that was never photographed. “Welcome home.” The film closes on both logos on a steel wall.
The storyboard as the client’s decision document
The storyboard held a timecode, camera note, scene content, voice-over line and a generated reference still for every shot. The client saw the film as a sequence of images before any motion existed, and could change order, copy and subject matter while changes still cost hours rather than shoot days.
How we separated the layers, then brought them back together
This was the heart of the work. Instead of demanding everything from a single prompt, we stabilised each layer on its own and combined them only afterwards.
Real people as a fixed cast. For each member of staff we first generated a studio portrait against a neutral backdrop: face, beard, hairline, build, in the black work shirt with the bar-and-shield on the chest. Those portraits went to the client for approval, and kept going back until colleagues recognised the person. Only then did the approved figure go into the scenes. When a face drifted in a scene, we reworked the face rather than regenerating the whole shot. We also learned to tell the model less. The more precisely we described a face in words, the more the model invented. The more plainly we said “this person, one to one from the photograph”, the better the likeness became.
The van from source. The van was referenced exclusively from its own photographs. We checked the livery against the original in every shot. Wherever HD KIEL or the bar-and-shield didn’t sit cleanly, the detail was corrected deliberately or placed from the original file, instead of waiting for a lucky generation.
Location first, then the figures. For each delivery location we first built an empty scene: harbour, alley, chalet, each with the van and its ramp down, but with no people and no motorcycle. We signed off that stage before any figure entered it. That way deviations in vehicle and surroundings were immediately visible, without a person in frame distracting from them. The real dealership was built exclusively from the reference photographs.
Motorcycles from source, riders bound to them. Each motorcycle was referenced from its own photograph and described by its distinguishing features: the orange tank, the chrome spokes with whitewall tyres, the matte black disc wheel. On the autobahn, every rider was fixed to his machine. With several riders in frame, models like to swap faces and motorcycles among them the moment the group briefly leaves the picture. So we defined rider and machine as one unit that does not come apart.
The logo isn’t generated, it’s placed. The bar-and-shield on the shirt, on the van and in the end card was never left to the model. Wherever it appears, it was placed from the original file and matched to the light of the scene.
Assemble first, animate second. Only once room, vehicle, figure and motorcycle were correct in the still did the keyframe get approved and animated. With direct text-to-video the model decides on rooms, faces and vehicles. With a keyframe-first approach the image is already signed off and the model only adds motion.
Motion you have to describe precisely
With a ramp, “they unload the motorcycle” isn’t enough. Who walks beside the machine holding the bars, who walks behind it supporting the weight, which way does the wheel roll, who is on the ground and who on the ramp. Every one of those questions had to be answered, or one of the staff would suddenly be sitting on the bike, or the machine would roll into the van instead of out of it. The same applied to the handshake in the Silberwolf courtyard: what should look natural has to be specified precisely, and gestures named too explicitly get exaggerated by the model. We learned what to leave out.
For the interior shots in the dealership we deliberately held the camera movement back. The less the camera travels through the space, the less the model can alter the counter, the lettering and the display.
Quality control on every clip
After each video generation the clip was checked against four questions. Is that still the same member of staff? Is the van still the van, with legible livery? Is the motorcycle still the right model, with the right rider on it? Is the location correct? Clips that failed a question were regenerated, or trimmed in the edit so the drift never becomes visible. This control step took more time than the generation itself.
Post-production as the connective tissue
The approved sequences were cut, timed to the voice-over, graded to a shared tonality and bound together with sound design: the phone at the counter, the van’s diesel, tyres on the ramp, five V-twins on the autobahn. Grading had a second job here. It put the staff, placed from portraits, into the same light as the scene around them, so that nobody reads as inserted but as photographed.
04. Production
A journey across Germany without the journey
A conventional production of the same scenes would have taken the van and a crew to Hamburg, into an old-town alley and up into the Alps, with three customers as actors, a night shoot at the harbour, an evening shoot in the mountains, motorcycles on a ramp under time pressure, an autobahn sequence with several riders and a scene in the Silberwolf courtyard in Brandenburg. Several shoot days in several places, with weather windows and permits. That is the kind of effort that normally shrinks a film about a delivery service down to a single location.
In a hybrid AI workflow the production work moves off the set and into reference preparation, concept, storyboard, image development, generation with per-scene quality control, and post. The result shows the whole of Germany, while the production costs a fraction of what the conventional route would have demanded.
There is a second effect that matters just as much day to day: changes stay possible at any point. An additional delivery location, a different machine, a new customer. Every adjustment is a new image, not a new shoot day.
05. The Result
A commercial with the real team, the real van and the real dealership
The result is a compact AI commercial that tells a service promise while showing the dealership as it is. The staff delivering the machines in the film are the staff customers get on the phone. The van is the van that pulls up outside the door. The film ends at the partner venue to which Harley-Davidson Kiel organises group rides, and the man shaking hands there is its owner. Both brands stand side by side at the close.
For the client that means a film for the website, social media, trade fairs and the showroom, set across the whole of Germany without anyone leaving Kiel, with a brand that holds up in every single frame.
06. Frequently Asked Questions
Can real employees appear in an AI commercial?
Yes. We work from several portrait photographs of each person, first build a character reference suitable for approval, and only place the person into scenes once it is signed off. Client approval is decisive here: whether a face is right is judged by the colleagues, not by us.
Can a real vehicle with livery be reproduced correctly?
Yes, if you show it rather than describe it. The van was referenced exclusively from its own photographs, and the livery was checked in every shot. Wherever lettering didn’t sit cleanly, it was placed from the original file.
How do scenes get made in places where nothing was ever shot?
Every location is first built as an empty stage and approved, then vehicle, figures and motorcycle go in. That is how a film can be produced across five locations in Germany without anyone travelling.
How do staff, vehicle and motorcycles stay consistent across scenes?
Through fixed references for each layer, established before any video is generated, and through checks on every keyframe and every clip. With several riders in frame, rider and machine are bound together so the model cannot swap them.
What does an AI commercial cost compared with a conventional shoot?
Shoot days, travel, actors, locations and permits all fall away. The effort sits in references, concept, image development, generation and post. A film like this is produced for a fraction of the cost of a conventional production. We calculate the specific scope individually, based on runtime and number of scenes.
Which businesses is this kind of AI commercial suited to?
Dealerships, retailers and service businesses that want to show their real team, their real vehicles and their real premises, and whose story plays out in more places than a shoot budget allows. It is especially worthwhile for service promises that span a whole delivery region.
Does AI replace post-production?
No. Edit, timing, voice-over, sound design and grading remain decisive. On this film, post was also where logos and livery were finalised and where the placed staff were bound to the light of their scenes.
Planning your own AI commercial?
We develop AI commercials from dramaturgy through references, storyboard and image development to finished post-production. Whether dealership, retailer, hotel or brand: tell us briefly what you have in mind and we’ll show you which images become possible with a hybrid AI workflow.