Backpropagation
Leads: Carl, Nadja
Ensemble: Angelica, Sonja, Peter, Max
Extended Synopsis
Blue activations pulse through a neural network in darkness. They pass from neuron to neuron, gathering into a prediction at the output neuron. Then orange signals race in the opposite direction, carrying the error backward. Plucked notes rise with the prediction and fall with the error. The network is learning, one training step after another, and with every step we draw closer to it, until we slip through the output neuron into the network itself. The strings come in as other networks emerge from the darkness, like stars, constellations and galaxies floating in a night sky. As the strings swell, the title BACKPROPAGATION appears, and a backward-facing drone view over a park fades in beneath it.
The park gradually comes into full view. Three girls race their bicycles along a gravel path. Nadja leads, with Sonja and Angelica close behind. The strings turn warmer and more romantic as we pick up their playful conversation.
SONJA …and he was leaning back so cutely just to get a glimpse of you!
NADJA You’re killing me with this stuff. You’re going to make it weird.
ANGELICA Nah, you can handle it. Just like you handle MedTech-1.
NADJA What does that even mean? I am going to ace this course. Ahhhh!
Their conversation fades beneath the strings, which swell into the film’s main theme as the drone rises higher and higher above the university. It looks down on the building, begins to spin, and then plunges toward it.
Inside, students on the Media Technology programme gather for a lecture on multimodal world models. Professor Sinsky begins with ordinary phone videos of a courtyard. He reconstructs the courtyard as Gaussian splats, creating a navigable spatial foundation from the recorded images. He moves through the reconstruction, pointing out how close some views already come to looking real.
The splat environment forms the base world. Sinsky then discusses how separately trained character models can be aligned with and inserted into it, with their bodies, voices and movements learned from additional recordings.
The students are given two weeks to create a navigable world model from a real location as the final exam of the course. After class, Carl, Peter and Max gather enough courage to ask Nadja, Sonja and Angelica if they’d like to do the group project together. The girls agree to join them.
The six walk back through the park together, the girls pushing their bikes.
ANGELICA It has to be somewhere real. And somewhere we can film in two weeks.
MAX We could do the park.
He points around them.
PETER Why would anyone visit the park in a world model? They can walk out the door.
ANGELICA The pond, then. Water is hard to reconstruct. That would at least be interesting.
PETER Half the course is going to do a pond. The other half is going to do the courtyard.
MAX (smiling) And they’ll pass just fine.
Peter walks past them and turns round, so that he is walking backwards in front of the whole group.
PETER Or. Hear me out.
Nobody stops walking. He keeps pace with them, backwards.
PETER Not a park. A world. The lake from the class trip. Three hours from the nearest road. We walk in and capture all of it: the trail, the forest, the shore, the water. And anyone on the course can walk the whole way there without getting out of their chair.
A pause. Sonja claps, slowly.
SONJA Beautiful.
MAX Three hours.
ANGELICA It was five last time.
NADJA That was the scenic route.
A look travels round the group. Nobody takes it up.
ANGELICA And there’s no electricity. Phones, laptops, the drone. What do we charge it all with?
PETER I’ll get solar panels.
MAX You don’t have solar panels.
PETER I’ll get them. Whatever it takes to bring the vision to life.
He is about to walk backwards into a lamp post. Max reaches out and steers him round it without looking.
The six settle on capturing a lake where they once camped together, several hours’ hike from the nearest road.
A few days later, the six of them set out. Angelica pilots the drone while Sonja, Carl and Nadja record fragments of the journey on their phones. Angelica sets the rules as they walk.
ANGELICA Twenty minutes per battery. Move slowly, overlap everything, and I guess keep people out of the shots when possible.
SONJA What if we’re very photogenic?
ANGELICA Especially then.
PETER Are we camping in the same spot as last time? Where the bear was?
MAX It was a fox.
PETER It was breathing like a bear.
MAX You’ve never heard a bear breathe.
PETER I have now.
SONJA When he first told it, it was a badger. It grows every time.
PETER And Carl went out there alone with a flashlight.
NADJA I went out with the flashlight. Carl came out after me with a frying pan.
CARL Someone had to go with you.
NADJA You stood behind me the whole time.
CARL Bears circle.
ANGELICA It was a fox.
The group walks on. Nadja drops back beside Carl.
NADJA Did you bring the pan this time?
CARL Obviously.
She smiles and walks on ahead.
A high, gliding shot follows the group as they emerge from the forest and the lake comes into view. Carl and Sonja break into a run toward the water. The lake lies almost perfectly still, surrounded by trees, with a narrow strip of shore where they pitch their tents. After resting, they spend several hours capturing the area. By evening, they feed the day’s footage into the model.
The result is crude but fascinating. The trail, forest and lake appear as a soft, navigable world. Because the students accidentally appeared in one another’s recordings, vague human shapes have also become part of the reconstruction. Peter finds a lumpy version of himself trapped halfway inside a tree. Sonja has three arms. Max appears and disappears depending on the viewing angle.
The group laughs until they are nearly crying. Angelica examines the scene more closely.
ANGELICA It’s treating us as part of the environment. We should keep the world layer fixed and add ourselves back as separate character models.
SONJA I object to being removed from the landscape.
They save this first version and decide to integrate themselves properly the next day.
In the evening, they sit around the fire. Someone has left a phone recording beside them to start capturing voices for their characters. It lies on a flat stone at the edge of the firelight with its red dot on.
Peter attempts to explain how neural networks are trained and backpropagation. He holds up his sausage stick and uses it to show the route: forward, toward the flames, and then back.
PETER The network makes a decision. Then it measures the error, and the error travels backward through the network to reconfigure its mind, so it makes a better decision next time.
NADJA So it learns by regretting?
CARL Regret as a derivative.
NADJA But you still need something real to compare it with. Otherwise it’s just fantasizing.
MAX How many times does it have to be wrong?
PETER Millions.
MAX That’s a lot of regret.
PETER Think of all the baggage these models bring into every new conversation.
SONJA No wonder it apologises so much.
ANGELICA And sometimes it gets stuck. You keep training and the error doesn’t move. It just sits there, flat, for ages. A plateau. It looks like the model has learned everything it’s ever going to learn. And then, if you keep going anyway — bang. It suddenly gets it.
CARL How do you know if it’s a plateau, or if it’s just finished?
ANGELICA You don’t. You keep going to find out.
MAX That’s me! I plateaued like crazy in Calculus, first semester. Then — bang — I got everything in the last two days and nailed a D.
Angelica laughs a little. Then she looks at the fire.
ANGELICA Peter. Your sausage is on fire.
Peter pulls his stick out of the flames. The sausage is black.
Sonja asks everyone increasingly inappropriate questions because, she claims, the model requires “emotional diversity.” She holds the phone out like a microphone.
The last shot of the night is a distant drone view: six small figures beside a moonlit lake, their voices too faint to make out. The fire dies down, and the image dissolves into the same landscape at dawn.
The following morning, green light comes through the wall of Carl and Peter’s tent. Peter is asleep on his back. Carl lies awake, already in his jumper, listening. Outside: the scrape of a spoon in a pan, Nadja and Sonja talking quietly. Nadja laughs.
PETER (eyes closed) Bring me a coffee.
CARL I’m not getting up yet.
Peter opens one eye, looks at Carl’s jumper, and closes it again.
Carl unzips the tent. Mist lies on the lake. He walks barefoot over the wet grass and finds Nadja crouched beside the stove, trying to turn a pancake with a spoon. Sonja sits beside her, in a sleeping bag pulled up to her armpits.
CARL That’s my pan.
NADJA You brought it for me.
Carl doesn’t argue. He pours himself a coffee and sits down.
Nadja points across the lake.
NADJA When we leave, we should take the path on the other side. It’s shorter.
CARL That’s where we got lost last time.
NADJA We missed a turn. If we take the turn, it’s shorter.
CARL You were very sure last time too.
NADJA I know where the turn is now.
He gives her a doubtful look. She laughs. Sonja interjects:
SONJA Tell him about when you got your high-school teacher to call the police.
NADJA I got three people lost on a school trip once. We were supposed to meet our teacher at this museum after lunch. I’d been there before, so I said I’d take them.
CARL Fair enough.
NADJA With my parents. When I was six.
Carl pauses.
CARL Well. They don’t usually move museums.
NADJA Exactly. That’s what I thought. I remembered these big steps outside. So I picked a street that looked familiar and started walking. Kept telling them we were nearly there.
CARL Once they’re following you, you can’t start hesitating. You’ve pointed. People are depending on you.
NADJA Thank you. I wasn’t going to let them down by wavering. One of the girls kept asking if I was sure, which wasn’t helping.
CARL Had she been to the museum?
NADJA No.
CARL Right. Then I’m not sure what she was contributing.
Nadja smiles and hands him the spoon.
NADJA So we kept going. I was still looking for the steps. And when I saw the burger place where we’d had lunch, I steered us into an alley before anyone noticed.
CARL Why were they still following you?
NADJA My leadership abilities… and the fun along the way. We popped into shops, watched street performers and looked at the architecture. And every time we got it wrong, I knew another way not to go. Like that backpropagation thing. You have to do the forward pass first.
Carl and Sonja laugh and nod.
SONJA You need something to correct.
NADJA By the time we finally found the museum, our teacher was on the phone to the police. Unfortunately, she didn’t know about backpropagation.
CARL If only she’d known.
NADJA She’d have congratulated me. So anyway, we definitely need to take that shorter path on the way back.
CARL Definitely. We know about backpropagation.
Carl laughs into his coffee. Sonja reaches between them and lifts the pan off the stove. They both look down at the blackened pancake.
CARL That one’s mine.
Later, they record one another walking, turning and making different expressions, combining the new footage with the voices captured around the fire.
When it is Sonja’s turn to film Max, he stands on a patch of grass with the lake behind him. The character app speaks from her phone in a calm, synthetic voice.
THE APP Please turn slowly to your left.
Max turns slowly to his left, very seriously, and stops.
THE APP Please show surprise.
Max raises his eyebrows a little.
SONJA Is that surprise?
MAX That’s surprise.
THE APP Expression not detected. Please try again.
Max raises his eyebrows a little, exactly as before.
SONJA Max.
MAX I’m surprised. I’m very surprised.
THE APP Please show joy.
Max’s face doesn’t change.
SONJA Joy, Max.
MAX That was joy.
THE APP Expression not detected.
MAX It’s subtle.
SONJA Think of something nice.
MAX Okay.
SONJA Food.
MAX I’m thinking about food.
His face doesn’t change.
SONJA A dog. Skiing. Your mum.
MAX A dog is nice.
Behind Sonja, further along the shore, Peter is carrying one of his solar panels down to the water. He steps on a wet rock, slips, and sits down in the lake, holding the panel high above his head so it stays dry.
Max bursts out laughing. It is a real laugh, and he can’t stop.
SONJA What?
She glances over her shoulder, then quickly back.
SONJA Don’t move! Keep doing that!
Max is bent over, laughing. Sonja keeps filming.
When it is Carl’s turn to capture Nadja, the others are busy further down the shore. She turns slowly in front of his phone as the software instructs, then holds still, facing the lens. Carl looks up from the screen. Nadja looks back at him and smiles. Neither of them says anything. The phone keeps recording.
Using Sinsky’s recommended character software, they train the six models separately and align them with the reconstructed world.
That evening, they test the more complete model. It can now simulate recognisable versions of all six students. The result is state-of-the-art, but still looks generated if one looks close enough.
Peter immediately uses the avatars to recreate scenes from The Notebook, where Carl takes Nadja rowing in the lake together with a million generated ducks. This amuses the group, although Carl and Nadja suddenly become self-conscious about standing so close together.
Peter then asks the model to recreate their arrival at the lake. An aerial image emerges. The six students walk out of the forest toward the water while the virtual camera glides above and around them. It is beautiful, immersive and only slightly stilted.
The shot is almost identical to the arrival shot the audience saw earlier in the film. For the first time, the viewer is forced to wonder whether a scene previously accepted as reality was already a reconstruction.
This feeling is reinforced by their conversation. Max points to Carl on the screen.
MAX The drone was dead by then, but I remember you and Sonja running ahead just like that.
NADJA I filmed them on my phone! It must have reconstructed it.
PETER Still a nice shot.
Over the course of their stay, cameras remain around them, but the assignment gradually recedes behind the experience itself. They swim, cook over the fire, explore the shore and talk late into the night.
Their conversations wander between jokes, plans and the project. They discuss whether generated images will eventually become more common than recorded ones, whether people will care where an experience came from if it feels real, and whether a model trained on someone’s messages, expressions and recordings could meaningfully know that person.
Many scenes are captured in long shots, including seamless transitions from within a tent (gimbal) to flying outside (drone), giving the scenes a beautiful almost artificial quality (the gimbal is attached to the drone during the shot)
They sit around the fire again at night and talk about the summer ahead. Peter has applied to several graphics and AI start-ups. Max wants to find seasonal work somewhere outdoors. Sonja refuses to make plans until she knows whether she has passed her exams. Carl has arranged a placement in the university lab. Their faces move in and out of darkness as the flames rise and fall.
Nadja says she has been thinking about making a documentary.
CARL About what?
NADJA I don’t know. Go somewhere, meet people, make something real.
CARL You don’t know what the film would be about?
NADJA I’d probably have to go there first.
Later, Carl and Nadja sit apart from the group beside the water. The lake holds the last light of the sky while their friends’ voices drift toward them from the fire, along with a slow pop song playing on someone’s phone.
NADJA It’s beautiful out here.
CARL (looking at her) Yeah.
NADJA But cold.
CARL Yeah. Should’ve brought a blanket.
Nadja looks at him for a moment. Carl glances back toward the fire.
They sit silently for a while. The others call to them from the fire. They linger a moment longer before running back.
After everyone goes to sleep, the film abruptly leaves the lake.
A spinning aerial shot looks down on an unfamiliar apartment building in the city centre at night. The camera descends along its exterior, passes through a window and moves down a dim, anonymous corridor. It approaches a closed apartment door, and the screen goes black.
The students wake beside the lake the following morning and prepare their submission. They decide that their avatars should present the group project and write the dialogue for them.
Their digital doubles stand beside the virtual lake as the generated camera moves through the reconstruction.
SONJA’S AVATAR We reconstructed the trail and lake from the phone and drone footage captured during the weekend.
MAX’S AVATAR The environment uses Gaussian splats as its spatial foundation. The six character models were trained separately and then aligned to move inside it.
PETER’S AVATAR None of this would have been possible without my solar panels.
The virtual camera turns toward the panels beside their tents.
PETER’S AVATAR Which I have also preserved in the model.
SONJA’S AVATAR Peter insisted.
The result remains imperfect, but it is surprisingly convincing.
PETER And this is the worst it will ever be.
With the submission finished, they close the laptop. For the rest of the day, the six of them hike, swim and enjoy the place without a deadline hanging over them.
Toward evening, Carl and Nadja sit alone at the water’s edge.
NADJA You look like you want to say something.
Carl meets her eyes.
CARL I’d follow you to the wrong museum.
Nadja smiles. She gets up, runs into the lake and calls for Carl to follow. He does. They swim beyond the reflections of the trees and float together in the quiet water.
As the sun approaches the horizon, they kiss.
Then the film returns to the unfamiliar apartment building.
It follows the same corridor, passes through the same door and enters the apartment. Someone is sitting before a large computer screen, watching the lake scene we have just seen.
The camera circles him and reveals Carl, perhaps slightly older, just a year or two. His hair is shorter. He has the beginnings of a beard.
On the screen, the old university project has been reopened inside a new generation of world-model software. In the year or two since the lake weekend, the technology has advanced further. The avatars can now be regenerated as photorealistic people with stable faces, convincing voices, natural movements and expressions, and consistent behaviour.
Carl pauses the image of himself and Nadja in the lake. He adjusts the time of day.
CARL Make it slightly later.
The scene renders again.
The computer interface disappears, and the film returns to the lake in full immersion. Carl and Nadja climb out of the water hand in hand. The sunset is a little darker now. They return to the fire.
Later, they sit wrapped in the same blanket while their friends talk around them. Nadja rests against him. The camera slowly pulls away as the remaining light disappears over the lake.
An older Nadja walks through the city beneath falling snow. She enters the same apartment building, follows the same corridor and opens Carl’s door. He is waiting inside.
They cook dinner together, laugh and talk about their work. Their conversations carry the rhythm of the lake weekend but have an almost frictionless quality. Nadja understands what Carl means before he finishes speaking.
The following morning, Carl leaves for work. His life outside the apartment is competent and colourless. He spends his days in front of screens reviewing code and evaluation metrics.
On the way home, Carl nearly walks into Max. Max is older too, but still recognisably Max. He is on his way home from work as well. Carl is looking at his phone.
MAX Carl!
CARL Max. Sorry, I wasn’t looking.
MAX What’s so interesting?
CARL Test results.
MAX Did you pass?
CARL The model did. Mostly.
MAX What is it this time?
CARL People. It makes people. Faces, voices, that sort of thing.
MAX Remember when Sonja had three arms?
CARL Now it can generate the rest of her life. These last couple of months, you can barely tell the difference.
MAX My mum keeps sending me videos of animals rescuing people. A dog pulling a kid out of a river.
CARL Does she know?
MAX I haven’t told her.
Carl laughs.
MAX As Peter said, this is the worst it will ever be.
CARL He was right.
MAX He still says none of it would have been possible without his solar panels.
CARL They charged one phone.
MAX I had dinner with Angelica yesterday. She asked about you.
CARL What did you tell her?
MAX That you’re fine.
CARL I am fine.
MAX That’s what I told her. Do you know what Nadja’s doing now?
CARL I think she’s still working as a journalist. Video, that sort of thing. But other than that, I’m not really sure anymore.
MAX Did you ever manage to charm her?
Carl gives a small, embarrassed laugh.
CARL I guess I don’t even know.
The answer makes it apparent that the scenes of Carl and Nadja living together were not from his present life. They were another scene generated by the model. Carl never acted on what passed between them at the lake.
Max studies him for a moment, sensing more than Carl has said. Then he makes an offer.
MAX Angelica and I are taking the night train to Hamburg next Friday. Two friends are coming. Come with us.
CARL What’s the plan?
MAX Hamburg first.
CARL And after Hamburg?
MAX We’ll find out.
CARL We’ve got a release that week.
MAX On a Friday night?
CARL I’ll see.
MAX Sure.
Max walks on. After a few steps he turns round.
MAX Ten past ten! Under the departure board!
Carl watches him go. Then he looks back down at his phone.
The story now turns to the real Nadja in the present, with a different hairstyle than at the lake and in Carl’s simulations.
She works as a video journalist at a newsroom. Several protests have recently taken place around the city, and a set of viral clips appear to show a large gathering already underway outside a suburban station.
Nadja compares the clips. The station and surrounding buildings look remarkably convincing. Her colleague, a camera operator, pulls up a chair beside her.
CAMERA OPERATOR Who posted them?
NADJA Four accounts. Two of them made this week.
CAMERA OPERATOR It looks real.
NADJA They all look real.
She opens archived street-view imagery of the square to check the location. As she rotates through the panorama, she notices a cyclist in a yellow jacket with a blue box over the rear wheel.
The same cyclist is visible behind the crowd in one of the viral clips—in the exact same posture, with the same wheel angle and folds in his jacket.
CAMERA OPERATOR He’s in both.
NADJA He isn’t at the protest. He’s in the map.
CAMERA OPERATOR When’s the map from?
NADJA (checking) The street view is from over a year ago. (to herself as much as to him) They used the street view as the world. Then they put the crowd in on top. And he came with the world.
CAMERA OPERATOR So it’s fake.
NADJA It’s fake.
She sits back.
NADJA There’s probably nobody there at all. Someone made it at their kitchen table.
CAMERA OPERATOR I’ll tell the desk.
NADJA Wait. Let’s call someone there.
By now it is early evening. Her colleague finds a small shop facing the square in the panorama, looks up its number and calls on speakerphone. Nadja does the talking.
SHOPKEEPER Hello?
NADJA Hi, sorry to bother you. I’m a journalist. Is there a protest outside your shop?
SHOPKEEPER A protest? Hold on.
Footsteps. A door opens, and the noise of the street comes through the speaker.
SHOPKEEPER Well, there are a few people out there now. Some of them have signs.
NADJA How many?
SHOPKEEPER Twenty. Thirty, maybe. More are coming out of the station.
NADJA Oh, they’re coming in now?
SHOPKEEPER Looks like it. A woman showed me one of those videos at lunch and asked where the protest was. There was nothing. It was completely calm.
NADJA When did people start arriving?
SHOPKEEPER Not long ago. I’ve been in the back most of the afternoon. Do you want to see? I can film it for you.
NADJA Yes, please. Just from the door is fine.
SHOPKEEPER Give me a minute.
They end the call. The camera operator looks at the crowd frozen on Nadja’s screen.
CAMERA OPERATOR They’re turning up because of this?
NADJA And now they’ll film it themselves.
Her phone buzzes. The shopkeeper’s video shows the square from his doorway: a crowd with signs, many of them holding up their phones, filming. Nadja puts it beside the viral clip. The real gathering is starting to resemble the generated one.
CAMERA OPERATOR I’ll get the car.
By nightfall, the protest depicted in the original clips has become real.
Nadja returns to the archived street view. She looks at the cyclist left inside the generated world again. It reminds her of the figures accidentally baked into their primitive lake reconstruction.
She searches for the old university project and opens the first saved reconstruction. Peter is still trapped halfway inside a tree. Sonja still has three arms. Nadja smiles, then continues into the original phone recordings and finds the conversation beside the fire.
PETER, ON THE ORIGINAL RECORDING The network makes a decision. Then it measures the error, and the error travels backward through the network to reconfigure its mind, so it makes a better decision next time.
NADJA, ON THE ORIGINAL RECORDING So it learns by regretting?
CARL, ON THE ORIGINAL RECORDING Regret as a derivative.
Everyone laughs.
Adult Nadja laughs quietly too.
The image holds for a moment on Carl beside the fire.
Cut to the same recording playing on Carl’s monitor.
Carl has returned to his apartment. Without the generated Nadja inside it, the place looks subtly different. He lets the campfire recording continue, then searches through the rest of the original lake archive.
The image quality is reduced and the tone less warm. Nothing is framed quite correctly.
Yet much of what Carl generated was built from something real. The conversations are almost the same, word for word. Nadja made the same jokes. She looked at him in many of the same ways. In the capture he recorded of her, she is not quite looking into the lens. She is smiling at him.
Carl finds a recording from the last evening, made by a phone propped against a rock on the shore. He and Nadja sit at the water’s edge, in the same place and the same light as in his version.
NADJA You look like you want to say something.
Carl hesitates.
CARL I think the model turned out really well.
NADJA (curious) That’s what you wanted to say?
CARL The character alignment, especially. If we’d had one more day for the voices—
Max and Peter come running down the shore, Peter with a phone in his hand. They race past Nadja and Carl and plunge into the lake. Sonja and Angelica follow, shouting after them. Nadja laughs and runs in with the others. Carl joins them.
The recording continues as all six swim, splash one another and compete to reach a floating branch.
Carl lets the recording continue. Nadja moves out of frame. His younger self is still there, laughing with the others.
Day becomes night outside the apartment. Images of the city are intercut with the lake as it is today, silent and empty.
Carl opens a rail booking page. He selects Friday’s night train to Hamburg and buys a ticket.
The generated Nadja enters behind Carl. She places her hands on his shoulders, removes his headphones and draws him to his feet. The slow pop song from the lake is playing.
They dance.
Carl looks at her. She is perfect in all the ways a person can only be perfect when someone else controls the prediction.
CARL I have to say goodbye to you now.
NADJA You don’t have to.
CARL I need my forward pass, remember?
The generated Nadja smiles, then looks at him with sadness.
NADJA Then stay until the song ends.
The music rises. They continue dancing as the apartment around them gradually fades into darkness.
When the song ends, Carl is alone at his desk.
A few days later, Carl enters the central station carrying a backpack. Max is waiting beneath the departure board with Angelica and two people Carl has never met. Max walks toward Carl, surprised and pleased.
CARL Where are we going after Hamburg?
MAX We still haven’t decided.
ANGELICA There are several deeply incompatible opinions.
Carl looks through the open train doors.
CARL Good.
Before they board, Angelica gathers the travellers together for a quick photograph. Angelica posts the photograph in the old university group chat.
ANGELICA — MESSAGE Look who actually showed up.
Nadja’s phone lights up. She sees Carl beneath the departure board with his backpack over one shoulder. She studies the photograph for a moment, then smiles.
The lake archive is still among her recently opened files. Nadja returns to it and finds another recording from the final evening. The file is just over twenty minutes long. It is an uninterrupted recording of the drone’s camera feed, with flight information visible along the edges of the image.
At the beginning, Peter stands at the edge of the frame holding the controller. He places it on the grass and runs down to join the others.
The six friends make a half-hearted attempt to pose for the drone. Nadja and Carl exchange a quick look—small, genuine and unresolved. Peter throws something at Max. Max charges after him. Sonja attempts to address the drone and is interrupted by Angelica. Within seconds, the entire group is laughing and running toward the water.
The clip continues far longer than Nadja expects. She looks at the timeline. Most of the continuous recording is still ahead of her. Nadja drags the playhead forward.
The image jumps through fragments of the evening. At four minutes, everyone is in the water. At eight, Peter and Max are wrestling over a branch. At twelve, the group lies on the grass while Sonja speaks animatedly and nobody seems to be listening. At sixteen, Carl and Angelica carry towels back from the tents. At eighteen, Nadja runs across the shore with Max chasing her. Nadja stops scrolling near the final minutes and presses play. The six younger friends continue playing beside the lake in the final evening light.
At the station, the train doors close.
The train begins to move.
The camera approaches Nadja’s monitor. The surrounding room disappears until the drone recording fills the screen completely.
Peter and Max chase one another along the shore. Angelica splashes Sonja, who immediately retaliates. Carl and Nadja pass through the frame together before separating again into the movement of the group. The theme music swells.
A low-battery warning appears.
RETURN TO HOME flashes over the image.
The drone slowly ascends and pans across the landscape, clouds and trees flowing by as it returns to its landing spot. It tilts down toward the grass and descends. As it touches the ground, we cut to black.
Now hiring