OLM Digital says SHIAGEDO adapts to a production from three hand-painted reference frames.
Why it matters: The experimental tool could reduce finishing work in 2D animation, but the reported performance figures lack detailed testing conditions. Human review remains necessary, and no commercial availability has been announced in the reviewed materials.
- OLM Digital says SHIAGEDO learns a production's visual style in about 90 seconds and processes each later frame in about 15 seconds.
- The three inputs are hand-painted reference frames used to adapt the system to a specific production, not the entire training basis for a general model.
- IMAGICA GROUP reported about 80% accuracy and a roughly 30% reduction in painting-related work in earlier testing, but the reviewed materials do not specify the task, dataset, sample size, evaluation date or human baseline.
- OLM developed SHIAGEDO through ANIMINS, a 15-month project under Japan's GENIAC program.
OLM Digital, the Japanese animation and visual-effects company associated with the Pokémon anime franchise, has developed an experimental coloring system for 2D animation, Variety reported.
Known as SHIAGEDO, the system is designed for the finishing stage, when colorists apply prescribed colors to line-art frames. OLM says three hand-painted reference frames can adapt the system to the visual style and palette of the production being processed. They are production-specific reference inputs, not evidence that the underlying general model was trained from only three frames.
OLM Digital R&D supervisor Tatsuo Yotsukura told Variety that the initial adaptation takes about 90 seconds and that the system takes roughly 15 seconds to process each subsequent frame. Those are company-reported performance figures.
The work was part of ANIMINS, a 15-month research project led by OLM Digital under Japan's GENIAC program, overseen by the Ministry of Economy, Trade and Industry with NEDO support. The broader project covered key animation, in-between animation, finishing and character drawing, METI said.
OLM describes SHIAGEDO as a production-specific assistant rather than a general image generator. Project records say commercial anime works were converted into research datasets with permission, while the underlying dataset was not planned for public release, official records show.
IMAGICA GROUP reported earlier testing results of about 80% accuracy in following color instructions and a roughly 30% reduction in painting-related work, but that report does not provide the task definition, dataset, sample size, evaluation date or a comparison against a human baseline. The figures should therefore be treated as company-reported results, not independently verified benchmarks. Every generated frame still requires human checking.
Yotsukura said the system is “strictly speaking” not generative AI. The reviewed project records and cited reporting contain no announcement of a commercial launch, pricing, public model or general availability date.
By the numbers
- 90 seconds - company-reported time for SHIAGEDO's initial production-style adaptation
- 80% - company-reported accuracy in following color instructions; testing conditions were not disclosed in the reviewed materials
- 30% - company-reported reduction in painting-related work; no human baseline was specified
Yes, but: The performance figures come from OLM or an industry participant and lack disclosed testing details, including the dataset, sample size, evaluation date and human comparison. Human review remains required.