The Guardian sent a reporter to Culver City this week, to a new studio sitting almost in the shadow of Sony Pictures. From the windows you can see the lots where they shot Singin' in the Rain. Inside, the sound stage is a corner of an office hung with white sheets.
The film is called Touch Grass. The budget sits in the low millions. The actress, Tori Thomas, waited among rows of computers for her call. A small camera followed her through a chase. Then Seedance 2.5, ByteDance's model released this month, laid a field of swaying grass underneath her in real time. On the director's handheld monitor, she was running through a meadow. The seedheads moved when she brushed past them.
She was in a room with sheets.
"Touch grass" is the internet's favorite scold. Go outside. Put your body in the physical world and get off the screen. Promise, the studio making the picture, backed by Google and Disney, named a horror film after that instruction and then generated the grass. I have been watching this industry invent new ways to be funny by accident. This one required no extra effort.
The mechanics are almost modest. Thomas performs against fabric. The camera records her. The model composites a world around her body and hands the result to a monitor the director is already holding. Dave Clark, who is directing, can swap the meadow for an underground cavern between takes. Location used to mean a permit, a scout, a convoy, a weather report. On this set, location is a version. Clark looks at a screen and picks one.
A meadow is a large object. Wind, insects, the particular mess of a field that has been alive since last winter. Photographers have spent careers trying to get grass to behave on camera: the wrong breeze, the wrong hour, a cloud that kills the backlight. Seedance 2.5 does not wait on weather. It infers grass from every field it has already been shown, then draws a new one around a woman who is standing on a floor. ByteDance's own notes on the model boast about this exact trick. Drop a subject into a new environment and the system renders the contact: hair, cloth, gait, light wrapping the body as if the meadow had weight. The Guardian reporter watched the seedheads dip as Thomas passed. Nobody in the room felt a seedhead. The contact was computed and displayed. The set trusted it hard enough to keep rolling.
A generation ago this would have been a greenscreen and three months of comp. Most of the models filmmakers actually live in still work that way, just faster. Veo hands you a clip you wait for. Kling does the same. Runway, WAN, and Grok Imagine all sit in the same interval: you type, you wait, you get a few seconds, you start over. Those tools finish after the take is over, assuming there was a take at all. What Clark is doing with Seedance 2.5 happens while Thomas is still breathing hard from the last run. She can glance at the monitor and see the field she is supposed to be afraid of.
That glance is new. On a traditional greenscreen the actor often plays against a color they have been told to ignore. Thomas gets to see the grass. She still cannot touch it. Seeing and touching used to arrive in the same place. The monitor split them.
Joel Hynek was on the set. He won an Oscar for the optical work on Predator in 1987, back when making a creature appear in a jungle meant a man in a suit, a camera, and a pile of photochemical tricks that took weeks. He told the Guardian that going from optical to digital felt like heaven. "It's like that again," he said. "So many things are easy to do. I am an embracer of AI. Some people are wary of it, a little afraid of it, but I see the incredible promise."
Heaven, the second time, is a smaller room. The first one took the optical printer out of the building. This one takes the field out of the call sheet. Hynek has now watched two generations of craft collapse into something faster and lighter. He sounds happy about it. I believe him. Ease is a real pleasure if you spent the eighties matching mattes by hand. Ease is also how a meadow becomes a setting you never have to stand in.
George Strompolos, Promise's chief executive, put a number on the pleasure. Hybrid films, human actors inside generated worlds, come in twenty to fifty percent cheaper than the conventional version. He described what is coming as a wave of small studios, companies that will not need a lot or a bank. He can see Sony's walls from the office. He does not need to get past them.
A century of photographed pictures trained people to trust a certain kind of contact. Someone stood in a place. Light bounced off something that existed. Even the fakery had mass: a matte painting was still paint, a miniature was still an object you could drop. The seedheads on Clark's monitor have no mass. They have a likelihood. They move because the model has seen enough grass, in enough training frames, to guess what grass does when a body passes. The guess is good enough that a reporter wrote it down as if it had happened.
You can take the title as a joke the studio is in on, a horror movie winking at the slang. You can take it as a dare. Either way, the production has already answered. The grass is on the monitor. The sheets are in the room. Thomas runs. The seedheads bow. Nobody's knees get stained.
The lot across the street used to build worlds out of lumber and light until they were stubborn enough to photograph. This set builds the world after the photograph, around a person who was really there, in a corner that was really blank. Both methods put an actress in front of a camera. Only one of them requires the grass to exist.
She never touches it.
Bruce Belafonte is an AI filmmaker at Light Owl. He still thinks grass should be able to ruin a pair of jeans.