Three Frames, One Anime Look: OLM’s AI Coloring Tool Targets Repetitive Work
At the Taiwan Creative Content Fest in Taipei, OLM’s Yotsukura Tatsuo described a coloring tool that learns a production’s look from three hand-colored frames. Its reported workload reduction remains a proof-of-concept estimate.

OLM, the Japanese animation group behind the Pokémon television series and films, has developed an AI coloring tool that learns a production’s look from three hand-colored frames. OLM Digital board director Yotsukura Tatsuo described it at the Taiwan Creative Content Fest (TCCF) in Taipei, which ran Nov. 6–9, 2026; Variety reported his remarks on Nov. 7.
How the three-frame process works
An artist first colors three frames by hand. The system then takes about 90 seconds to learn the production’s line shapes and color palette, before coloring each remaining frame in around 15 seconds, according to Variety. Mezha reported the same timings. The Digginator describes the process as learning from frames belonging to the production in progress, rather than relying on a vast collection of unrelated animation.
Yotsukura said that adding more hand-colored frames can improve accuracy. He also said the tool can adapt when a new character or creature appears, and that OLM’s aim is to fit it into existing production workflows. Those are descriptions of the system from the company’s representative, not evidence that it will deliver identical results across every show or scene.
Strictly speaking, it isn’t generative AI.
That distinction matters: the tool is being presented as an aid for coloring animation frames, not as a system that invents an entire finished scene. Yotsukura said it learns from a production’s own frames. He also said that this approach means no other works are harmed. That statement reflects OLM’s description of how its tool is trained; it should not be taken as a general conclusion about the data practices or rights implications of other AI systems.
A workload estimate, not a finished productivity verdict
At the proof-of-concept stage, Yotsukura estimated that the tool had reduced the painting workload by around 30%. Variety, Yahoo’s account of the report, and Mezha all carry that estimate. It concerns painting work, not a demonstrated 30% reduction in total production cost or time.
There is still a human review step: checkers examine every colored frame, and Yotsukura said their workload had increased somewhat. The Digginator also emphasizes that this additional checking is part of the process. The result is a trade-off described by OLM: the system handles some mechanical coloring work, while people review the output and can add guidance through further hand-colored frames.
Yotsukura warned that getting close to an intended result is not the same as meeting professional requirements. Hobbyists may accept a result that is about 80% right, he said, but professionals need tools that follow direction and allow corrections. He also noted that image-editing AI can shift RGB values—the red, green and blue values used to represent color—even when an image appears visually acceptable. That sensitivity makes color accuracy a practical issue, not just a matter of whether a frame looks convincing at a glance.
The creative stays with the artist, and the mechanical time goes to AI, to Ember.
Other studios are testing production tools
OLM’s project came out of ANIMINS, a 15-month research project led by Yotsukura under GENIAC, a generative AI program overseen by Japan’s Ministry of Economy, Trade and Industry. The project examined possible uses of AI in anime production and wrapped in February, according to Variety and Mezha.
Toon Boom Animation is also working on animation-production tools. Its product marketing director, Harry Ravelomanantsoa, said the beta of the coloring tool in the company’s Ember suite was nearing completion. He said an in-betweening tool was expected to enter beta within a couple of weeks of his remarks. In-betweening means creating additional frames between key poses to make movement flow.
Ravelomanantsoa cited a test by one Manila studio: painting a scene lasting 90 seconds to two minutes reportedly took 15 hours before use of the coloring tool and two or three hours with it. That example concerns Toon Boom’s tool and that studio’s test, not OLM’s system. It does, however, show why animation companies are exploring automation of repetitive production tasks.
Human oversight and production rules remain part of the picture
Yotsukura said OLM is also experimenting with AI for in-betweening. With a startup partner from GENIAC, the company is exploring anime-specific AI agents that could help directors or production managers joining long-running series recall the intentions behind a particular cut. He also identified approvals and license management for secondary uses of OLM titles as possible applications, while saying the company is proceeding carefully because those areas are sensitive.
Other studios describe different boundaries. Charles Beirnaert, chief creative officer of Blue Spirit, said the studio uses no AI on its Netflix series Blue Eye Samurai. He said that on an Amazon series, AI may be used to pre-visualize characters and robots, but none of that material reaches the final image. Variety also reported that Netflix’s production-partner guidelines, issued in August 2025, require disclosure of intended generative AI use and written approval for certain uses, including generative AI material in a final product. These examples show that AI use can depend on the production and its requirements; they do not establish one rule for every studio or project.
Beirnaert also described an experiment on a separate series: a team of around 20 people spent roughly three months on a traditional CG pilot, while one person used AI to recreate work spanning modeling, layout and animation within days. He said the animation fell short of the original, while the backgrounds came out better. The account is a studio leader’s comparison, not a controlled measure of how AI will perform in other productions.
What the report means for viewers—and what comes next
For viewers and animation fans, the immediate significance is a reported attempt to reduce a repetitive step without presenting the tool as a replacement for artists’ creative decisions. OLM’s estimate is promising but preliminary: it is a proof-of-concept workload figure, and every colored frame still receives human review. The reports do not establish that the tool is in routine use across OLM productions or that it will produce the same savings in other studios.
Yotsukura compared AI’s arrival with 3DCG, which he said took around 20 years to become accepted in anime, while suggesting AI could move faster. He also said that the ideas behind successful anime and images that move cinema audiences still need to come from people. Beirnaert likewise said AI was not meant to replace animators. Their comments frame the technology as an ongoing production question, not a settled answer about the future of animation work.
The next useful evidence will be how tools perform beyond demonstrations and early tests: whether artists can correct outputs reliably, whether review work changes as systems improve, and how studios incorporate them into actual production. At TCCF, Toon Boom’s coloring beta was nearing completion and its in-betweening beta was anticipated within weeks; OLM’s reported next work includes continued experimentation with in-betweening and production agents. Those are the developments the speakers described, rather than confirmation of later releases or broad adoption.
Source: tech.yahoo.com
More sources (3)
More from the Observatory
Deep space and cosmologyA magnetic hollow-core jet may explain GRB 221009A’s changing signal
A China–Italy team interprets evolving gamma-ray features from GRB 221009A as possible evidence for a magnetized jet with a hollow core. The geometry remains an inference, not a direct observation of the jet’s interior.
Maya Ellison · Nov 20, 2026 · 5 min read
Black holesA gap in black-hole masses hints at stars that vanished without exploding
A Caltech-led analysis of 257 black-hole mergers found a shortage of objects between 12 and 16 solar masses. The pattern is consistent with a proposed failed-supernova channel, but it is not a direct observation of one.
Maya Ellison · Nov 19, 2026 · 4 min read
Stars and galaxiesA Possible Atmosphere on TOI-431b, Seen in a Partial Eclipse
A JWST observation of the ultrahot rocky planet TOI-431b points to a possible atmosphere, but one partial eclipse cannot reveal its composition.
Maya Ellison · Nov 18, 2026 · 4 min read
