What’s the difference between this and pose transfer?

Pose transfer uses 2D keypoints; this workflow uses per-frame depth maps. Depth preserves camera motion and scene geometry (parallax, relative distances), so movement feels more 3D and faithful to the original shot.

Can I keep the original background or subject identity?

By design, no. The depth pass removes pixel identity and color, and the Seedance 2.0 synthesis rebuilds visuals from the reference image. If you need the original background, composite later in your editor using the new render and the source plate.

Why does the output flicker or distort on fast motion?

Depth quality drops with motion blur or low light. Use cleaner source footage (higher shutter speed, more light), keep aspect ratio consistent between source and output, and avoid aggressive cropping. Matching FPS between VHS_LoadVideo and SaveVideo also helps stability.

How do I keep timing and audio perfectly in sync?

Keep the frame range and FPS consistent from VHS_LoadVideo through SaveVideo, and connect VHS_LoadVideo’s audio output to SaveVideo’s audio input. This preserves the original timing and soundtrack in the final render.

Workflow

Depth Motion Capture with Seedance 2.0

Feed a reference video through depth extraction to strip away everything but structure — pure spatial motion, no pixels, no color, no identity. That depth sequence becomes the skeleton. Hand the skeleton to Seedance 2.0 as your motion reference, then drop in a reference image to define the look. The depth map dictates how things move; the image dictates what they are. Camera moves, body motion, and scene geometry carry over intact while the visual identity is fully swapped. Dial in duration, resolution, and aspect ratio to match your target format. Route the original audio straight through to the final output so the new footage stays locked to the source timing.

جرب على Comfy Cloud تحميل سير العمل

الفئات

توليد الفيديو video video-edit video-generation style-reference style-transfer pose-transfer motion-control character-reference character

النماذج

seedance-2-0

أنشأ بواسطة

Purz 1 day ago

Seedance 2.0 Depth Dancer turns any source clip into a motion “skeleton” by extracting a depth map sequence, then rebuilds the visuals from a single reference image. In this workflow, VHS_LoadVideo ingests your reference video (and audio), the custom Seedance 2.0 node (4871bd59-7f4e-475a-a872-bf5457365402) converts each frame to depth to capture camera moves, body motion, and scene geometry, and then synthesizes new frames guided by a look you define with LoadImage. Finally, SaveVideo assembles the generated frames and can pass through the original soundtrack to keep everything in sync.

Technically, the depth pass strips away color and identity, leaving only spatial structure over time. Seedance 2.0 uses that per-frame depth as the motion driver while the reference image sets the subject’s appearance and style. You control duration (frame range), resolution, and aspect ratio in the Seedance 2.0 node to match your target format, and you connect the audio from VHS_LoadVideo to SaveVideo so timing remains locked to the source. A MarkdownNote is included in-graph to document settings and the recommended routing.

FAQ

الأسئلة الشائعة

عرض جميع سير العمل