VACE Two Video Bridge Prep

Everything else in this pack is built around iterating over a clip list. VACE Two Video Bridge Prep is the direct path instead: drop in two native VIDEO inputs, tell it how many frames to replace on each side, and it builds a clean left/right bridge window for generating one smooth seam. The README calls this the preferred route for "editor/NLE use" - no loop, no clip list, no intermediate files. Just left clip, right clip, one transition.

How it works

You give it left_video and right_video (native Comfy VIDEO objects - load them with VACELoadVideoFromPath or your usual loader). It reads the frames and FPS out of both, then:

• left_replace_frames (default 16) - frames replaced from the end of the left video
• right_replace_frames (default 16) - frames replaced from the start of the right video

The visible bridge span is left_replace_frames + right_replace_frames. But here's where the node earns its keep: Wan generation lengths follow a 4n + 1 pattern (17, 33, 49, 81...), and a 32-frame visible bridge isn't a valid window. So Prep pads the seam with masked frames up to the next valid length and returns wan_length, padding_start, and padding_frames telling you exactly where that hidden padding is. The README's example: a 32-frame visible bridge becomes a 33-frame Wan control window, and VACE Two Video Bridge Assemble trims it back down to 32. You never see the padding; it just means your "ask for 32, get 32" request actually works.

edge_blend_frames (default 4) marks frames at each outer bridge edge that stay anchored to the original footage - used both as guidance for the model and later as blend anchors when Assemble re-mixes the generated edges back into the source.

Inputs and outputs that matter

• left_video, right_video (VIDEO) - the two clips
• left_replace_frames, right_replace_frames, edge_blend_frames - the three numbers above
• resize_strategy - passthrough (default) derives size from matching inputs; explicit_resize stretches both videos to width/height (multiples of 16)
• fps (optional) - an FPS override; leave disconnected to derive and validate FPS from both videos (mismatched FPS errors)
• width, height (optional) - only used when resize_strategy is explicit_resize
• debug - console prints of all the derived numbers

The outputs you'll actually touch: control_video + control_mask (into your VACE control nodes), left_kept + right_kept (the un-replaced clip bodies), bridge_source (the original frames of the bridge span, used for edge blending), and wan_length/bridge_frames/padding_*/edge_frames to wire into the Assemble node. fps and bit_depth come out for encoding, and color_reference_frame is a single frame (the last non-gray left anchor) for color matching - feed it to something like KJNodes' ColorMatch if you're chasing consistent color.

Installing it

Part of the comfy-vace-automation pack, no extra pip dependencies:

cd ComfyUI/custom_nodes
git clone https://github.com/EnviralDesign/comfy-vace-automation

Restart ComfyUI, or install via ComfyUI Manager (search "comfy-vace-automation").

When it bites you

• Mismatched inputs. With passthrough, both videos must share resolution and channels; with mismatched FPS you'll get an error unless you set the fps override. Read the error - it names both sides.
• explicit_resize without width/height errors immediately, and the sizes must be multiples of 16.
• Replace counts bigger than the video. left_replace_frames can't exceed the left video's length, same for the right. The error tells you the limit.
• Bridge span too long. The visible maximum is 80 frames (the Wan window tops out at 81). If you need a longer transition, split the work - this node is designed around the 4n+1 window, not around breaking it.

For the VACE side, remember this pack doesn't bundle the inference nodes - the README points at external packs like KJNodes for extras such as WanVideoNAG and ColorMatch. Prep hands off standard control video + mask, so it slots into whatever Wan VACE setup you already run.