Why Jurassic Park’s Dinosaurs Still Look Real

Generated video is the newest kind of digital effect, and a brand buying it faces a question film crews have argued about for decades: what should be real in front of the camera, and what can be made afterward? The best-known answer is more than thirty years old. Jurassic Park still convinces audiences, and much of the reason is how little of it relied on computer graphics.

Filmmakers call the physical work done on set practical effects: puppets, rigs, rain machines, models. Digital effects are added in post, and most films mix the two. AI video models are the newest layer on the digital side, able to generate whole shots, people and places included.

Fifteen minutes of dinosaurs, six of them digital

Jurassic Park runs just over two hours. According to ILM, the effects house that built its digital creatures, the finished film holds six minutes of computer-generated dinosaurs and nine minutes of practical puppet work. Peter Hodges, a visual effects lecturer at the University of South Wales, puts the total at 15 minutes on screen, with the nine minutes belonging to Stan Winston’s animatronics.

The digital share almost didn’t happen. The creatures were planned as stop-motion, and ILM says Steven Spielberg approved the switch to computer graphics after seeing tests, with less than a year left before release. Even then, the digital animals kept a physical source: Phil Tippett’s stop-motion team moved armatures fitted with motion encoders, and that data drove the computer models. Writing for Smithsonian, Adam Bargteil notes that Spielberg “approached computer-generated sequences with caution.”

The two techniques share the same scenes. In the T. rex attack, Hodges writes, animatronic close-ups come first, and full-height digital shots then deliver the animal’s threat and power. The puppet earns belief at arm’s length. The digital dinosaur does what no puppet could, walking and running in full view.

That split maps onto generated video. A model can produce what a brand could never build or schedule: an aerial approach to a resort at dawn, a mountain road in the season you missed. Close shots are where viewers inspect, and where generation is weakest at holding a label or a hand steady. Real at close range and generated in the wide is a sound default for an AI project.

Where Jurassic Park’s dinosaurs came from

Minutes of dinosaurs on screen, by technique

Minutes of dinosaurs on screen in Jurassic Park, by technique

Technique	Minutes
Animatronics and puppets (Stan Winston)	9
Computer-generated animation (ILM)	6
Total	15

Weight and light come free with a real object

A physical object answers to physics whether the crew likes it or not. The T. rex was so big that Winston’s team had to modify its own building, raising the roof and enlarging the door, ABC News reported. The script added a tropical storm, and Stan Winston Studios had not waterproofed its full-size T. rex. In SlashFilm’s account, puppeteer John Rosengrant recalls the foam-rubber skin soaking up water and throwing off the hydraulics, so the animal shuddered. Between shots the crew mopped it with towels, and a night shift dried it with blowers and hair dryers.

The same rain ran down the puppet and the actors, under the same lights. Weight, wet skin and reflections were recorded by the camera instead of reconstructed. American Cinematographer quoted Winston before release promising a “beautiful, seamless mix of technologies.” The digital shots had to rebuild by hand everything the puppet got for free, and the cut between them asks the audience to accept both as one animal.

Generated video starts on the digital side of that line. A model paints light and motion from patterns in its training data; it does not simulate mass or optics. In a 2025 benchmark called Physics-IQ, researchers led by Saman Motamed tested video models on fluid dynamics, optics and other physical principles and concluded that “visual realism does not imply physical understanding.” Grounding a generation in a real photograph helps: when a model animates a real still of your product in its real setting, light, surfaces and scale come from the photo instead of being invented.

Actors need something to look at

One of the film’s most remembered scenes was added on set. SYFY’s account says Spielberg was moved watching Winston’s sick triceratops breathe, asked Sam Neill to lay his body on its belly, and got more than he asked for from Laura Dern: “She actually cried. I didn’t ask for the tears.”

Where nothing was there, the cast worked blind. Dern told Business Insider in 2025 that on day one the actors had to react together to a faint sound that might be the T. rex. With the camera rolling, Spielberg roared through a megaphone, and Richard Attenborough turned to her, unconvinced. “We had no idea,” she said.

The puppets were physical enough to hurt. ABC News recounts that during the kitchen scene, a raptor pushed on wheels hit young Joseph Mazzello in the face with its metal claw. A puppet sits at a real height and distance, so every look lands on something. When a creature is added later, the crew has to mark where it will stand, and a gaze a few degrees off reads as someone staring past it.

Generated video breaks eyelines both ways. A generated presenter may look near the product rather than at it, and the gaze can drift within one clip. A filmed person reacting to a generated element has the megaphone problem; the fix is a physical mark on set, then a generation built from that real plate. Where a person handles the product, keep both real.

“No CGI” usually means CGI nobody noticed

Practical effects have become a marketing line. Creative Bloq reported that Tom Cruise said of Top Gun: Maverick that “everything you see is for real,” while about 400 artists sit in its visual effects credits. VFX Voice reports 2,400 visual effects shots in the film, including jets digitally re-skinned, and 1,300 in Barbie, a figure surfaced by visual effects supervisor Jonas Ussing, whose video series takes “no CGI” claims apart shot by shot.

The flying was real: the producers paid the US Navy for flight hours in F/A-18 Super Hornets, with cameras aimed at actors in the back seat. The F-14 and the enemy Su-57 were made entirely by computer, and footage of a single jet was used to create several on screen. Why sell it as practical? A visual effects executive quoted by VFX Voice says studios don’t want to undercut the actors’ marketing power. Christopher Nolan once compared his talk about CG to an actress discussing her Botox.

The lesson for AI video runs both ways. Audiences reward a digital layer they cannot see, and these films hid theirs under real faces, aircraft and light; a generated shot needs the same anchor. And “real footage, no AI” over composited shots is Cruise’s line, which anyone can now take apart frame by frame.

What to keep real, and what to ask

In most brand videos the practical layer is short: the product, the hands that use it, any face making a claim, and the location when it is what you sell. Skies, wide shots, seasons and crowds can be generated around them. Before approving an AI video project, ask:

• Which shots are filmed, which are generated, and which combine both?
• What real photograph or plate is each generated shot built from?
• Is the product ever generated, or always filmed or photographed and composited?
• On composites, who checks that light direction, shadows and eyelines match the real layer?
• Will every claim made about the video, such as “shot on location,” still be true after post?