深度视频控制

使用源剪辑作为移动深度图，使新世界继承其相机路径、空间层次和主体体积。

Control behavior

场景替换

相机连续性

体积风格化

1–15s · fixed 24 fps · final price follows measured GPU compute

Describe the replacement

OUTPUT SIZEMAX DURATIONCONTROL STRENGTH 1.00SEED

Extract DEPTH from my normal videoTurn this off only when the upload is already a prepared control pass.I accept the MiniMax H3 Community License and its use restrictions. This model is unavailable in the US, EU, UK, and South Korea.

Estimated priceFinal price is 1.2× measured compute~144 credits$1.44

Output

Your new world follows the source clip’s depth and camera motion.

场景替换

相机连续性

体积风格化

One union model, six control workflows

Switch between Canny, Depth, HED, MLSD, Pose, and masked video inpainting without changing checkpoints. Upload ordinary footage and ManifoldGen prepares the matching control pass.

Try Pose