Industries / Media, arts and recreation / Media and entertainment
Animation and visual effects
Making pictures move one frame at a time by drawing, posing models or computing images, and combining filmed and made images so that they read as one shot.
7
Milestones achieved, from 1892 to 2019.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1892
Drawings move on screen
Émile Reynaud's Théâtre Optique, patented in 1888, projected long strips of hand-painted pictures onto a large screen over a fixed painted background. From 28 October 1892 paying audiences at the Musée Grévin in Paris watched his "luminous pantomimes," with music and sound effects, three years before the Lumière brothers' first public screening. France's national film center, the CNC, notes that World Animation Day has marked the date each year since 2002.
Credit
Émile Reynaud
1908
One drawing per frame
On 17 August 1908 the caricaturist Émile Cohl showed Fantasmagorie at the Théâtre du Gymnase in Paris. He made it by photographing his drawings one frame at a time with a film camera, a use the camera was not built for, and the CNC calls it the first film made this way. Cohl made more than 300 films by 1937, some mixing drawings, objects and live action.
Credit
Émile Cohl (Émile Courtet)
1933
A model ape stars
For RKO's King Kong, released on 2 March 1933 at Radio City Music Hall and the RKO Roxy in New York, Willis O'Brien moved 18-inch and 24-inch models of the ape a little between frames, a method now called stop motion, and combined them with actors through rear projection, his own "miniature projection," which placed actors inside small-scale sets, and composites (images layered from separate pieces of film). Kong's fur rippled on screen where the animators' fingers had touched it.
Credit
Willis O'Brien; Marcel Delgado, who built the models; Merian C. Cooper and Ernest B. Schoedsack, producers and directors
1937
A feature-length cartoon
Walt Disney's Snow White and the Seven Dwarfs premiered on 21 December 1937 at the Carthay Circle Theater in Hollywood. More than 750 artists worked on it for three years, at a cost of $1.4 million. Disney's own encyclopedia, D23, calls it "the first animated feature film."
Credit
Walt Disney Productions; supervising director David Hand
1995
Made entirely in computers
Pixar's Toy Story, made with Disney and released on 22 November 1995, is described by the Academy of Motion Picture Arts and Sciences as "the first computer-animated feature film." Its programmers were still building the software while the story was being written, and its director, John Lasseter, received a Special Achievement Oscar at the 68th Academy Awards. It was also the first animated feature nominated for Best Original Screenplay.
Credit
Pixar Animation Studios (John Lasseter, Ed Catmull); Walt Disney Pictures
2013
Oscar winner goes bankrupt
Life of Pi won four Oscars, including visual effects, for a computer-made tiger, orangutan, meerkats and ocean; Rhythm & Hues did much of the work, and more than 600 effects artists worked on the film. The effects supervisor, Bill Westenhofer, tried to raise the company's money trouble in his acceptance speech; by March 2013 Rhythm & Hues was bankrupt and up for auction, NPR reported. Studios paid effects companies a fixed fee per shot, however long the work took and however many changes the director asked for, and three studios had lent the company $20 million to finish their films.
LessonA fixed price per shot puts the cost of every director's change on the effects company, and even an award-winning one can fail under it.
Credit
Rhythm & Hues Studios; Bill Westenhofer
2019
Backgrounds on a wall
For the first season of The Mandalorian, which premiered when the Disney+ streaming service launched on 12 November 2019, Industrial Light & Magic surrounded a stage 75 feet across with a curved LED wall 20 feet high and 180 feet around, plus an LED ceiling, showing 3D backgrounds drawn in real time by Epic Games' Unreal Engine. The camera was tracked, so the background shifted in perspective as it moved, which American Cinematographer said set the method apart from earlier projection systems, and the show's creator Jon Favreau told the magazine in February 2020 that "a majority of the shots were done completely in camera."
Credit
Industrial Light & Magic (StageCraft); Epic Games; Jon Favreau; cinematographers Greig Fraser and Barry Idoine
Watch for
Generative AI in final shots
Netflix co-CEO Ted Sarandos said that a building collapse in the Argentine series The Eternaut, made with Netflix's Eyeline Studios, was "the very first [generative] AI final footage" in a Netflix original. He said the sequence was finished 10 times faster than with traditional visual-effects tools and that its cost would otherwise not have been feasible on the show's budget.
Achieved whenAn outside party, such as an effects company, union or auditor, publishes the time and cost of comparable shots made both ways, or such shots appear across many productions with published counts.
Claimed by
Netflix, July 2025 (IBC report of 23 July 2025)
An AI-made animated feature
Critterz, an animated feature produced by AGC International, Vertigo Films and Native Foreign with OpenAI's tools, was set to premiere at the Cannes Film Festival in May 2026 on a budget under $30 million and a schedule of about nine months, against about three years for a comparable traditional animated film. It missed the festival. OpenAI shut down Sora, the video model the film had used for sequences, in March 2026, and its web and app versions went dark on 26 April; the producers showed first-look footage to buyers at the Cannes market instead, The Next Web reported.
Achieved whenThe finished film is released to audiences and an outside account confirms its budget, schedule and how much of it was generated.
Claimed by
OpenAI and the producers, reported by The Wall Street Journal in September 2025
Open
01
Score AI roto against artists
A public test of AI rotoscoping. Rotoscoping, or roto, means tracing the outline of a person or object in every frame of a shot so that it can be separated from its background, and it is one of the slowest jobs in visual effects. Have professional roto artists make reference outlines for a set of openly licensed live-action shots, run current AI tools on the same shots, and publish how many shots each tool delivers that an artist would accept unchanged or with minor fixes, and how many minutes of fixing each needs.
The briefCredit · unclaimed02
Grade AI in-betweens blind
A blind test of AI in-betweening for hand-drawn animation. A key animator draws the main poses of a movement, and in-betweeners draw the frames that come between them. Commission key drawings, have both human in-betweeners and AI tools fill in the frames, and have working animation checkers grade every frame without knowing who made it.
The briefCredit · unclaimed
Open 01 · the brief
Score AI roto against artists
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of AI rotoscoping. Rotoscoping, or roto, means tracing the outline of a person or object in every frame of a shot so that it can be separated from its background, and it is one of the slowest jobs in visual effects. Have professional roto artists make reference outlines for a set of openly licensed live-action shots, run current AI tools on the same shots, and publish how many shots each tool delivers that an artist would accept unchanged or with minor fixes, and how many minutes of fixing each needs.
Why it is still open
Intel and the stop-motion studio Laika reported in 2021 that their AI roto tool produced 26% of shots usable with no adjustment and another 47% with minor adjustments, but only for seams and features on puppet faces, tested on frames from one film. This research found no public test of AI roto tools on live-action effects shots graded by roto artists.
Built with what exists
The raw footage of Tears of Steel, a 2012 live-action effects short whose shots the Blender Foundation released under a Creative Commons Attribution license; SAM 2, Meta's open-source model that cuts objects out of video; Blender or Nuke for compositing; the three grades Intel and Laika used (no adjustment, minor adjustment, redo).
First step
Download ten Tears of Steel shots with hair and motion blur, run SAM 2 on them, and have one roto artist grade the results on the three grades.
Done looks like
40 shots with hair, motion blur or smoke, each with an artist-made reference; for each of at least three tools, the share of shots two independent roto artists accept unchanged or with minor fixes, and the median minutes of fixing, published together with all the outlines.
Open 02 · the brief
Grade AI in-betweens blind
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A blind test of AI in-betweening for hand-drawn animation. A key animator draws the main poses of a movement, and in-betweeners draw the frames that come between them. Commission key drawings, have both human in-betweeners and AI tools fill in the frames, and have working animation checkers grade every frame without knowing who made it.
Why it is still open
A January 2025 survey of generative AI for hand-drawn animation lists in-betweening among the tasks AI is starting to take over and says problems of visual consistency and style remain. Tencent's ToonComposer (August 2025) reports better results than earlier methods on its own benchmark built from human-drawn sketches. This research found no published blind test in which studio checkers grade AI and human in-betweens side by side.
Built with what exists
ToonCrafter, an open-source model that fills in frames between two cartoon drawings (Apache 2.0 license); ToonComposer, whose code Tencent ARC has published; OpenToonz, free and open-source 2D animation software; a drawing tablet.
First step
Commission ten key-drawing pairs from a freelance animator under a license that allows publication, and generate in-betweens for them with ToonCrafter.
Done looks like
50 pairs of key drawings, with three in-betweens each from humans and from each tool; two checkers grade every frame as pass, fix or redo; the share of passing frames and the minutes needed to fix the rest, published for humans and for each tool.