DeepGHS previews AnimeCSIP-v3, an artist-style recognition model built on AnimeDINOv3
bdsqlsz · x · 2026-10-26
DeepGHS previews AnimeCSIP-v3 (Contrastive anime Style Image Pre-Training), a downstream application of their AnimeDINOv3 model for artist-style recognition, retrieval and clustering. Demo charts show it judging whether two anime images share the same artist across different characters, with a prediction threshold of -0.582. Version 3 fixes hard pairs that tripped v2.1—sketch-to-color matching, same character in different styles—while some failure cases remain.
More from Multimodal
- Claude Code rebuilt its own ComfyUI pipeline to fix prop continuity in a 2-min AI film — Odd_Definition_9724 · 2026-10-26
- Gingerbread watercolor storybook clipart prompt shared with polished sample outputs — azed_ai · 2026-10-26
- Eleven v4 Turbo hits ~150ms first-chunk latency across 90+ languages — sven_ai · 2026-10-26
- Hailuo H3 Fast shows off anime PV chops: transformations, effects, fast-paced editing — Hailuo_AI · 2026-10-26
- HeyGen Video 1 and Voice 1 land on Sunra: one-pass dialogue, 39-language voice — toolstelegraph · 2026-10-26
- Fuzzy crayon-texture character art trend: fluffy glowing pastel style takes over X — OVolosin82152 · 2026-10-26