沐阳 H3 · 动作视频自动分段(内部)
Slicing the action video into per-segment chunks
- image
- 分段动作参考
Motion transfer in this pack works by slicing: your one action reference video gets carved into one piece per generated segment, and each segment follows the part that corresponds to its time position. 沐阳 H3 · 动作视频自动分段(内部) (H3ReferenceClip) is the node that does the carving. It's internal to the Director's motion-transfer path, so you'll rarely wire it by hand - but it's a clean, small idea, and knowing what it does explains half the Director's reference handling.
Given an image (a batch of frames, i.e. your reference video), a start_frame, and a frame_count, it returns the slice start_frame : start_frame + frame_count. The default frame_count is 125 - at 24fps that's just over 5 seconds, roughly one segment's worth on H3's time grid.
The interesting bit is the tail handling. If the slice runs off the end of the video - say the last segment asks for 125 frames and only 40 remain - the node pads by repeating the last available frame until the clip is full. That's deliberate: H3 expects each segment to line up with its frame grid, and a short reference slice would desync the timing. A repeated final frame is a defensible stand-in for "the action finished here," and it beats silently feeding a short clip into a pipeline that assumed a specific length.
Inputs
- image - the action reference video as an IMAGE batch.
- start_frame - 0-based, clamped into range.
- frame_count - how many frames to take, default 125.
Output: 分段动作参考 (IMAGE), the sliced clip.
Install and context
Pack install is the usual: ComfyUI Manager search "ComfyUI-MiniMaxH3-Myang", or git clone https://github.com/civilcoco/ComfyUI-MiniMaxH3-Myang into custom_nodes, restart. No extra Python deps.
You'll see this node appear inside the Director's expanded motion-transfer graph, usually in step with 沐阳 H3 · 动作音频自动分段 (its audio twin, slicing the soundtrack on the same frame range) and 沐阳 H3 · 分段素材替换 (which drops the slice into the media bundle under a stable @视频1). If a motion-transfer segment looks like it's re-treading the end of the reference video, this is the node whose clamping you're seeing. And the pack-wide reminder stands: H3's local weights are territory-restricted by community license (US/EU/UK/South Korea excluded) - check before you commit a long project.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — | |
| start_frame | INT | 00–1000000 | — |
| frame_count | INT | 1251–10000 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| 分段动作参考 | IMAGE | — |