depth-to-video test with MiniMax H3, fully local.
pipeline: I recorded a depth map of my hand (left) and built a style reference sheet in Ideogram 4, running locally on my MacBook (bottom). H3 uses the depth map for motion and the sheet for look. rendered on my old gaming PC, no cloud.
output on the right. finger articulation follows the depth pass closely, and the style transfer holds across the whole clip.
genuinely impressed with how far local video gen has come.