OLM reveals AI coloring tool trained on three hand-painted frames
The Pokémon anime studio says its proof-of-concept tool learns a show’s look in 90 seconds, while artists and checkers stay in control.
By Bianca Rossi · Entertainment Editor
2 min read
OLM’s AI coloring tool can learn an anime production’s visual rules from three hand-painted frames, according to OLM Digital board director Yotsukura Tatsuo. The Japanese group behind the Pokémon anime series and films disclosed the proof-of-concept system at the Taiwan Creative Content Fest in Taipei, Variety reported.
The workflow starts with an artist coloring three frames by hand. Yotsukura said the system then takes about 90 seconds to learn the production’s line shapes and color palette, before coloring subsequent frames in roughly 15 seconds each.
How does OLM’s AI coloring tool work?
The system is intended to use a show’s own artwork as its reference rather than a vast pool of outside training material. As artists color more frames, Yotsukura said, the output becomes more accurate. He also said it can adjust when a new character or creature enters a production.
That makes the tool a tightly guided coloring aid, not a hands-off animation factory. Artists establish the visual target, and human checkers still review every colored frame. Yotsukura said those reviewers may have somewhat more work even as painting work is reduced.
He said the tool is not, in the strict sense, generative AI. His point was that it learns from production-specific frames to apply an established look, rather than creating an image from a broad training set.
What has OLM reported so far?
OLM’s estimate remains preliminary. At the proof-of-concept stage, Yotsukura said the system had reduced painting workload by about 30%. That is an estimate from the studio, not evidence that the tool is in commercial use or deployed on Pokémon.
The research came from ANIMINS, a 15-month project led by Yotsukura under GENIAC, a generative-AI program overseen by Japan’s Ministry of Economy, Trade and Industry. The project concluded in February, according to Variety.
OLM’s stated goal is to fit the system into established anime production processes without disrupting them. The studio is also experimenting with AI for in-betweening, the work of creating drawings between key poses, but the coloring system is the specific use case Yotsukura described.
Why does human review remain part of the process?
Yotsukura said professional animators need tools that reliably follow their instructions and allow corrections. He also pointed to a color-control issue: image-editing AI can alter RGB values even where an image appears acceptable at first glance.
For now, OLM is presenting the three-frame system as a supervised production aid. Its reported speed comes with artists providing the references and people checking the results before a frame is cleared.
This story draws on original reporting from Variety.