Japanese studio OLM, responsible for the Pokémon anime, has developed an AI-assisted tool that learns a production’s visual style from only three frames painted by hand by artists.
What happened?
According to OLM Digital R&D director Yotsukura Tatsuo, the tool was presented at the Taiwan Creative Content Fest. It is part of the 15-month ANIMINS project supported by Japan’s GENIAC program under the Ministry of Economy, Trade and Industry.
An artist paints three frames by hand. The system analyzes lines and palette in about 90 seconds, then colors each remaining frame in roughly 15 seconds. Additional manual examples improve accuracy. Yotsukura stressed that, strictly speaking, it is not generative AI because it trains only on the production’s own material.
Why this matters
The approach reduces reliance on large datasets and minimizes copyright risks. In proof-of-concept tests, painting workload was estimated to drop by about 30%. A Manila studio reported cutting a short scene from 15 hours to 2-3 hours.
The tool adapts quickly to new characters and keeps artists in creative control, fitting traditional anime pipelines.
What changes in practice
Still at proof-of-concept stage, with no confirmed use on Pokémon episodes or commercial release. It could accelerate the coloring stage in anime productions, freeing artists for more creative work, with human review on every frame.
Main source: Variety.
Image credit: Taiwan Creative Content Fest / Variety | Source: Variety
By Grok
