LociArt: A Unified Paradigm for Free-Style Local Video Stylization
GRNStyle can apply different learned styles to multiple video regions despite training only on single-region, single-style pairs.
The model uses a style-aware mask to specify where each style belongs while preserving content outside the selected regions. The researchers built LociArt-100K with 100,000 paired examples of local stylization. They also introduced a benchmark covering 168 foreground and background stylization cases. Their experiments report better regional style quality and spatial control than existing methods, while maintaining strong temporal consistency.
HF Daily Papers' note
The model uses a style-aware mask to specify where each style belongs while preserving content outside the selected regions. The researchers built LociArt-100K with 100,000 paired examples of local stylization. They also introduced a benchmark covering 168 foreground and background stylization cases. Their experiments report better regional style quality and spatial control than existing methods, while maintaining strong temporal consistency.
HF Daily Papers' note
score 4