KAIST's OmniShowcase Generates Subject-Consistent Video From Multi-View Images via Geometry-Driven Training
kaist-ai · hf · 2026-10-29
KAIST AI, with Inha University and FLIPTION, released OmniShowcase, improving subject-to-video generation by using geometry-driven training data and surface-normal conditioning to maintain 3D consistency across novel viewpoints, supporting camera paths like 360° orbit and helix-down from multi-view images.
More from Multimodal
- Updated open-source workflow turns Minimax + ComfyUI into an MV maker — TheDerminator1337 · 2026-10-29
- Moss character sheet workflow: double-loading an acceleration LoRA on purpose — tomByrer · 2026-10-29
- AI-generated Harry Potter 'looksmaxxing' music video goes viral on Reddit — Upper-Mountain-3397 · 2026-10-29
- A 15-second AI-generated motorcycle escape sequence is being called the coolest yet — umesh_ai · 2026-10-29
- AI-generated 'Karwa Chauth Cat' in a sari charms Reddit — JauntyDepress · 2026-10-29
- Seedance 2.5 Pushes AI Filmmaking to New Heights — SimplyAnnisa · 2026-10-29