See-through auto-splits a single anime illustration into up to 23 layers for 2.5D PSD output

See-through, a research project that automatically decomposes a single static anime illustration into up to 23 semantic layers and outputs it as a 2.5D-manipulable PSD with hidden regions filled in, has been released as open source and is drawing attention. Officially titled "See-through: Single-image Layer Decomposition for Anime Characters," the preprint was posted to arXiv on February 3, 2026, and was conditionally accepted to SIGGRAPH 2026. Inference code is available on GitHub.