Pokemon Studio OLM Builds AI Coloring Tool Trained on Just Three Frames

OLM, the Japanese animation studio behind the "Pokemon" anime series and films, has built an AI coloring tool that learns a show's visual style from just three hand-painted frames. OLM Digital board director Yotsukura Tatsuo announced the development at the Taiwan Creative Content Fest (TCCF) in Taipei, running November 6–9 at the Taipei Nangang Exhibition Center.

The tool emerged from ANIMINS, a 15-month research project Yotsukura led under GENIAC, a generative AI program overseen by Japan's Ministry of Economy, Trade and Industry. The project, which examined how anime creators could use AI, concluded in February.

How does the tool actually work?

AI models normally demand vast training data that studios struggle to assemble. OLM inverted the problem. An artist colors three frames by hand. The system then learns that production's line shapes and color palette in about 90 seconds, and colors each remaining frame in roughly 15 seconds. The more frames an artist paints, the more accurate the system becomes.

"Strictly speaking, it isn't generative AI," Yotsukura said. Training only on a production's own frames means no other works are harmed, he added. When a new character or creature, such as a dragon, appears, the tool adapts immediately.

At proof-of-concept stage, Yotsukura estimated the tool has cut painting workload by around 30% — though checkers who review every colored frame now have somewhat more to do.

Toon Boom Animation, the Canadian maker of Harmony and Storyboard Pro, is attacking the same bottleneck. Product marketing director Harry Ravelomanantsoa said the beta of the coloring tool in Ember, its AI suite, is wrapping up, with in-betweening expected to enter beta within weeks. One Manila studio that tested the coloring tool cut painting time on a 90-second-to-two-minute scene from 15 hours to two or three.

"The creative stays with the artist, and the mechanical time goes to AI, to Ember," Ravelomanantsoa said.

What does this change for animators?

Yotsukura likened the gap between AI demos and production reality to gacha, the Japanese capsule-toy machines where each JPY500 ($3.30) turn may or may not yield the prize — roughly the cost of an AI generation credit. AI reaches close enough to 80% to raise expectations. "Raising it from there to 90 or 100% turns into work so hard it can make you heartsick," he said. Hobbyists may accept 80%; professional animators need tools that return their direction 100% of the time.

Anime is also highly sensitive to color, Yotsukura noted. AI tools that edit an image can shift its RGB values even when the result looks fine. "AI can only do what it has learned," he said, adding that the images that move cinema audiences and the never-before-seen ideas behind hit anime still have to come from people.

He compared AI's arrival to that of 3DCG, which took around 20 years to gain acceptance. "Twenty years ago, I kept rejecting it, saying, 'How could this ever be used in anime?' Now it's unthinkable for anime not to have 3DCG," he said. AI will move far faster, he argued, because its quality is already high at the start.

OLM employs around 600 staff across Japan, Malaysia, Thailand and the U.S. Its slate includes "Beyblade," "Hanakappa" and "The Apothecary Diaries." With a startup partner from GENIAC, OLM is exploring anime-specific AI agents to help new directors joining series that have run a decade or more, and to speed approvals and license management for secondary uses — the most revenue-critical part of the business, according to Yotsukura.

Where do Western studios stand?

Charles Beirnaert, chief creative officer of Blue Spirit, the French studio behind Netflix's "Blue Eye Samurai," said the studio uses no AI on that series. "We're not allowed to use AI for the Netflix projects," he said. Netflix's guidelines issued in August 2025 require partners to disclose intended generative AI use, with written approval needed for material in the final product. On an Amazon series, Blue Spirit may use AI only to pre-visualize characters and robots.

Elsewhere, Beirnaert's team tested AI on a greenlit CG series: one person recreated in days a pilot that around 20 people spent roughly three months making. "For me, AI is not meant to replace animators. The day every animator is replaced, I will change job and do something else," he said.

Asked about savings, Beirnaert estimated €1 million on an €8 million budget. Yotsukura said OLM, brought in only after budgets are fixed, sees AI as a way to gain creative freedom within them — and the question is shifting from how many staff AI replaces toward new shots AI tools make possible. "I think 'not using AI' is probably no longer an option from here on," he said.