Dance-video-trained model revealed: disoccluded mask inpainting drives the motion magic
pixlpa · x · 2026-10-18
Researcher pixlpa shares training details of their video model: one iteration was largely trained on improvisational dance videos. Motion is the focal point of the research, while image generation, regeneration, and inpainting of disoccluded masks are key to how these videos transform.
More from Multimodal
- Running Fish Audio Drama 3 on a 16GB AMD GPU: Red Curtain ports the 35B MoE that officially wants 48GB — fechyyy · 2026-10-18
- Kling 4.0 opened 48 hours ago: 7 wild examples with full prompts — minchoi · 2026-10-18
- AI-generated 'Motion Blur' fashion images stun with long-exposure look — OVolosin82152 · 2026-10-18
- AI-generated cryptid video wows Reddit with eerily realistic forest goblin — NailgunnerNorth · 2026-10-18
- AI-generated 'Tales from the Japan Open' video goes viral for its smoking dog — BlueMonStar · 2026-10-18
- fal is so fast someone built a generative filesystem on top of it — gorkem · 2026-10-18