VFX History.
Trick photography did not come after cinema: it was born with it, within a few months. From the 1895 stop trick to virtual production and AI, this course follows the thread of special effects (SFX) and visual effects (VFX) - illustrated with the collections of the National Cinema Museum in Turin and the backlots of Cinecittà.
SFX and VFX are not the same thing
Special effects (SFX) are made on set and captured live by the camera: makeup and prosthetics, pyrotechnics, rain and wind, animatronics, miniatures, trick sets, forced perspective. Visual effects (VFX) are made on the image: multiple exposures, mattes, keying, computer graphics, compositing. The border keeps moving - a miniature is a practical effect, but the moment it is combined with another image on an optical printer or in compositing, it becomes VFX. The history that follows is the story of this constant back-and-forth between the workshop and the lab, then between the set and the computer.
SFX and VFX are not the same thing
Special effects (SFX) are made on set and captured live by the camera: makeup and prosthetics, pyrotechnics, rain and wind, animatronics, miniatures, trick sets, forced perspective. Visual effects (VFX) are made on the image: multiple exposures, mattes, keying, computer graphics, compositing. The border keeps moving - a miniature is a practical effect, but the moment it is combined with another image on an optical printer or in compositing, it becomes VFX. The history that follows is the story of this constant back-and-forth between the workshop and the lab, then between the set and the computer.
Illusion before cinema
The whole mental toolkit of the trick film exists before the first screening. The magic lantern (Christiaan Huygens, c. 1659) projects hand-painted glass slides; its mechanical slides already animate moon phases or lever-driven figures, and Robertson's phantasmagoria shows (Paris, 1798) conjure ghosts onto smoke - projection was an art of the supernatural long before Méliès. In the 19th century, optical toys break movement down: Joseph Plateau's phenakistiscope (1832), William Horner's zoetrope (1834), Émile Reynaud's praxinoscope (1876), whose Théâtre optique projected drawn pantomimes from 1892. In parallel, the chronophotography of Eadweard Muybridge (1878) and Étienne-Jules Marey (1882) freezes motion frame by frame. Decompose, recompose, project: only film stock is missing.
The trick is born with cinema
The first trick even predates the Lumière brothers' first public screening (December 28, 1895). In August 1895, at Edison's, Alfred Clark shot The Execution of Mary, Queen of Scots with cameraman William Heise: the camera stops, the actress is swapped for a dummy, the camera rolls again, the axe falls. Editing did not exist yet, so the stop trick makes the cut - the first effect in history is an invisible splice. On the Lumière side, projecting Demolition of a Wall (1896) in reverse - the wall rebuilding itself - revealed that you could fake not just the shot, but the direction of time itself. Two founding discoveries: the image can lie about what happened, and about the order in which it happened.
Méliès, the magician
Georges Méliès, stage magician and director of the Robert-Houdin theatre, attended the first Lumière screening and understood at once that cinema was an instrument of magic. As the legend goes - he told it himself - a jammed camera at Place de l'Opéra revealed the stop trick to him: on playback, an omnibus turns into a hearse. From The Vanishing Lady (1896) onwards he transposed his stage tricks to the screen, then systematised the whole grammar of the trick film: multiple exposures on black backgrounds (The Four Troublesome Heads, 1898), mattes, dissolves, enlargement by tracking (The India Rubber Head, 1901), miniatures, stage machinery, and frame-by-frame hand colouring in Élisabeth Thuillier's workshops. His glass studio in Montreuil produced about 500 films, including A Trip to the Moon (1902), the first special-effects blockbuster. In the same years, Alice Guy - the first woman director in history (La Fée aux choux, 1896) - opened the fairy-tale vein at Gaumont.
The trick school - Porter, Blackton, Chomón, Cohl
The trick film steps out of filmed theatre. Edwin S. Porter combines storytelling and effects in The Great Train Robbery (1903): through the telegraph office window a train passes - a double exposure over a matte, the direct ancestor of matte painting and compositing; some prints were stencil-coloured by hand. His Dream of a Rarebit Fiend (1906) piles up delirious superimpositions. James Stuart Blackton animates chalk drawings frame by frame (Humorous Phases of Funny Faces, 1906), then objects (The Haunted Hotel, 1907): stop motion is born. The Spaniard Segundo de Chomón pushes it further (El Hotel eléctrico, 1908), before joining… the Itala Film studios of Turin, where he devised the effects and tracking shots of Cabiria (1914). Émile Cohl, finally, gives animation its first drawn character with Fantasmagorie (1908). From this lineage comes all of cel animation.
Mechanical trickery - sets, miniatures, stunts
The silent era perfects in-camera effects. Glass shots (sets painted on glass placed in front of the lens) and hanging miniatures complete partial sets - Chaplin would skate along a painted void in Modern Times (1936). Italy invents the monumental film: Cabiria (Pastrone, 1914), shot in Turin, established colossal sets and the tracking shot, and inspired Griffith; the heritage lives on at Cinecittà, opened in 1937, where sword-and-sandal epics are still shot. The great comics, meanwhile, cheat through precision: Buster Keaton literally walks into the screen in Sherlock Jr. (1924) - matches measured to the inch, sets swapped between camera stops - and in Safety Last! (1923) Harold Lloyd climbs a façade built above a rooftop, framed to hide the safety platform: forced perspective as a stunt.
Stop motion - from O'Brien to Harryhausen
Willis O'Brien turns stop motion into the art of the monster: The Lost World (1925), after Conan Doyle, puts animated dinosaurs opposite real actors, and King Kong (1933, Cooper & Schoedsack) weaves animation into live action through miniature rear projection and mattes - the monster becomes a performing character. His apprentice Ray Harryhausen takes over: an Oscar for Mighty Joe Young (1949), The Beast from 20,000 Fathoms (1953), then Dynamation - the animation sandwiched between a rear-projected background and a foreground matte - culminating in the skeleton fight of Jason and the Argonauts (1963): seven skeletons, four and a half months of animation for a few minutes of screen time. He closed the era with Clash of the Titans (1981). Stop motion lived on: in Bristol, Aardman has animated the clay of Wallace & Gromit since 1989, up to the Academy Award for Best Animated Feature for The Curse of the Were-Rabbit (2005); in the United States, Laika 3D-prints thousands of replacement faces for Coraline (2009) and Kubo and the Two Strings (2016); and Guillermo del Toro's Pinocchio (2022) won the Oscar in turn. On set, the mechanical successor is the animatronic - motorised creatures performing live.
The optical era - mattes, blue screens, keying
Meanwhile the lab invents compositing. Mattes and counter-mattes combine several takes on the same strip of film; the optical printer, perfected by Linwood Dunn on King Kong, lets shots be re-photographed and combined after the fact. The Williams process (1918) extracts subjects with a travelling matte - it is what “undresses” The Invisible Man (1933), shot in black velvet against black. The Dunning-Pomeroy process introduces colour-based extraction, and Petro Vlahos's sodium vapor process at Disney pulls off the composites of Mary Poppins (1964). The principle - isolating a subject with a colour absent from it - becomes the optical blue screen, then the digital green screen: the most universal tool of modern VFX.
2001, Star Wars and the ILM era
2001: A Space Odyssey (Kubrick, 1968) takes the optical effect to its summit: front projection for the Dawn of Man landscapes, Douglas Trumbull's slit-scan for the Star Gate, miniatures of documentary precision. Nine years later, Star Wars (1977) industrialises everything: George Lucas founds Industrial Light & Magic (1975), where John Dykstra builds the Dykstraflex - a camera with programmed, repeatable moves (motion control), finally allowing dozens of miniature elements to be composited into one moving shot. Next to this optical revolution, creatures remain handcrafted: costumes and masks, puppets (Yoda, performed by Frank Oz), and the animatronics of the Italian Carlo Rambaldi - the head of Giger's Alien (1979) and E.T. (1982), which earned him two of his three Oscars.
The digital turn
Digital enters in small steps: the robot's pixelated vision in Westworld (1973, the first computer-processed imagery in a feature), the first 3D in Futureworld (1976) - Ed Catmull's digitised hand -, the vector worlds of Tron (1982). At Lucasfilm, the Graphics Group builds the tools, is bought by Steve Jobs (1986) and becomes Pixar. ILM strings together proofs of concept: the water pseudopod of The Abyss (1989), the liquid-metal T-1000 of Terminator 2 (1991). The tipping point is Jurassic Park (1993): the CG tests by Mark Dippé and Steve Williams convince Spielberg to drop the go-motion planned by Phil Tippett - the photorealistic dinosaurs are digital in the wide shots, and animatronic (Stan Winston) in the close-ups. Thirty years on, that hybrid recipe is still the golden rule of photorealism.
Rotoscoping - tracing reality
Max Fleischer patents the rotoscope in 1917: project a film frame by frame behind frosted glass and trace the real movement - Koko the Clown is born from the gestures of his brother Dave, and Disney would use it as acting reference on Snow White (1937). Ralph Bakshi turns it into a style of its own: his animated Lord of the Rings (1978) and then Fire and Ice (1983, designed with Frank Frazetta) are almost entirely rotoscoped. For Fire and Ice, every fight is first performed by costumed actors on a bare stage, then traced and painted in Frazetta's style: one of the most accomplished examples of the technique, where the bodies keep the weight and momentum of real motion. Video games pick up the torch: Jordan Mechner traces his brother David's runs for Prince of Persia (1989), and Éric Chahi films his younger brother on video, overlays the footage on his Amiga through a genlock, and rotoscopes the animations of Another World (1991). Digital tools reinvent it as interpolation with Bob Sabiston's Rotoshop (Waking Life, 2001; A Scanner Darkly, 2006). Rotoscoping is the manual ancestor of two pillars of modern VFX: roto/matte extraction… and motion capture.
The music video as effects lab - Gondry & co
In the 1990s-2000s, the music video is where experimentation moves fastest. Michel Gondry replays all of Méliès with contemporary means: stop-motion Lego for The White Stripes' Fell in Love with a Girl (2002), an entire landscape synchronised to the music in The Chemical Brothers' Star Guitar (2002), a split-screen palindrome in Sugar Water (1996). His technical summit: Kylie Minogue's Come Into My World (2002) - a Milo motion control rig set in the middle of a Paris intersection repeats exactly the same move over four loops, and Olivier Gondry's compositing (rotoscoping, difference mattes) multiplies Kylie and the passers-by within the same shot. Chris Cunningham answers with the robot lovers of Björk's All Is Full of Love (1999). The Directors Label documentaries (“The Work of Director…”, 2003 - Gondry, Spike Jonze, Chris Cunningham) remain the best film school of modern trickery; Gondry carried this craftsmanship up to Eternal Sunshine of the Spotless Mind (2004). And the inheritance has moved to social media: Zach King and his “digital magic” (Vine, then TikTok and its tens of millions of followers) does pure Méliès - stop tricks, invisible cuts, mattes - on a smartphone, six seconds at a time. Trick photography no longer needs a studio: it has become close-up magic again.
The Lord of the Rings - old magic + digital
Peter Jackson's trilogy (2001-2003) summons every age of the trick film at once. To put Gandalf and the Hobbits in the same live shot, Weta reinvents forced perspective: the two actors play in the same frame but at different distances, on sliding platforms synchronised with the camera move - the size illusion holds even in a tracking shot, as in the Shire cart scene. Add scale doubles, sets and props built in two sizes, and the “bigatures” - miniatures so large the word no longer fits (Minas Tirith, Barad-dûr). On the digital side: Massive, Stephen Regelous's crowd software, animates the battles, and Gollum - played by Andy Serkis - brings performance capture into screen acting. Three visual effects Oscars in a row, including The Return of the King (2003).
Performance capture - from Avatar to Thanos and Dune
The mocap suit becomes standard: retroreflective markers, infrared cameras, a calibrated capture volume; Avatar (2009) adds facial capture via head-mounted cameras (HMC) and virtual-camera shooting - and becomes the biggest box-office hit in history. Andy Serkis extends Gollum's lineage with King Kong (2005), then Caesar in Planet of the Apes. Marvel pushes facial photorealism: Thanos in Infinity War and Endgame (2018-2019) is Josh Brolin's performance, transferred through machine learning (Digital Domain's Masquerade system) - and Endgame dethrones Avatar at the box office. At the opposite end of the spectacular, Dune (2021, 2024) practices “invisible” VFX: Denis Villeneuve and supervisor Paul Lambert (DNEG) replace green screens with sand-coloured screens so the composites marry the desert light - two films, two visual effects Oscars.
The digital pipeline - a factory of shots
Behind every shot of a modern blockbuster, an industrial chain. Previz sketches the shot in light 3D; matchmove reconstructs the real camera; roto and prep clean the plate. Modelling and texturing build characters and sets (Maya, 3ds Max, ZBrush, Blender); simulations - destruction, fluids, fire, cloth, crowds - run in Houdini; lighting renders the passes in Arnold, RenderMan or V-Ray. Compositing assembles everything in Nuke - node-based, the studio standard - or Flame for finishing, while After Effects rules motion design; colour grading (DaVinci Resolve, Baselight) finally unifies the light across every shot. At the scale of Avengers: Endgame: about 2,500 VFX shots, spread across ILM, Weta Digital, DNEG, Framestore, Cinesite and Digital Domain - the VFX credit roll is longer than that of every on-set trade combined.
Virtual production - the set comes back to the stage
The latest reversal: virtual production. The LED walls of ILM's StageCraft on The Mandalorian (2019) display the digital set behind the actors, live, with parallax tracked to the camera - the rear projection of King Kong (1933) reinvented in real time, with the bonus of true reflections and true set light on the actors' faces. VFX physically returns to the stage where it was born.
AI - from deepfakes to generated shots
Machine learning slipped into VFX through the back door, the one of thankless tasks. In late 2017, deepfakes appeared online: neural networks learn a face from thousands of images and place it onto another. Studios took hold of it: Digital Domain transferred Josh Brolin's performance onto Thanos with machine learning (2018), and after Luke Skywalker's de-aging in The Mandalorian (2020), ILM hired the YouTuber Shamook, whose deepfake version had convinced more. In software, AI settled in as an assistant: After Effects' Rotobrush extracts subjects, Nuke's CopyCat (2021) learns a fix from a few frames and repeats it across the whole shot, DaVinci Resolve's Magic Mask isolates a face, Photoshop's generative fill (Firefly, 2023) invents an image's edges. On Everything Everywhere All at Once (2022), a tiny visual effects team used Runway's tools for rotoscoping.
Then came generation. In 2022, DALL·E 2, Midjourney and Stable Diffusion made images from a sentence; in 2023, Runway moved to video (Gen-1, Gen-2); in 2024, OpenAI's Sora, Google's Veo and Kuaishou's Kling generated shots several seconds long, and Veo 3 (2025) added sound. The best gauge of this acceleration is a joke: in March 2023, “Will Smith eating spaghetti”, generated with ModelScope, made the internet laugh with its melting faces; two years later, the same request gives, with Veo 3, an almost believable shot, sound included. Node-based interfaces like ComfyUI, the logic of Nuke applied to diffusion models, and platforms like Higgsfield chain these models into real production pipelines. In cinema, Robert Zemeckis's Here (2024) de-ages and ages its actors in real time, on set, with Metaphysic's tools. The profession is negotiating its safeguards: the 2023 strikes by American writers and actors were partly about AI. A chapter being written right now, which asks the 1895 question all over again: what is an image when you can no longer tell how it was made?
Sources & places Museo Nazionale del Cinema, Turin (Mole Antonelliana) · Cinecittà si Mostra, Rome · films: Archive.org (public domain) · R. Rickitt, Special Effects - The History and Technique · personal photos, Xavier Reyé, 2025.
Five uncaptioned loops to identify, then five questions on the history of effects.
