Collects multiple VIDEO inputs and derives IMAGE clip batches plus shared FPS
Loops across an ordered in-memory clip list without saving intermediate videos
Optionally adds one final tail-to-head loop pass after the last normal seam
Carries the accumulated joined clip forward between iterations
Prepares native VACE control frames and masks for a single seam
Assembles normal seams and final loop seams without duplicating or extending the first clip body
Prepares and assembles a final tail-to-head VACE loop pass for one clip
Loads native Comfy VIDEO objects from absolute or project-relative paths
Prepares and assembles direct left/right video bridges for NLE-style workflows
Provides a standalone seed INT node with native Comfy seed-widget behavior

Loading videos from project paths

Use VACE Load Video From Path when the video file lives outside ComfyUI's input directory.

Inputs:

path: absolute video path, or a relative video path
base_directory: optional base directory for relative project paths

If path is relative, the node checks base_directory first when provided, then ComfyUI's input directory, then the Comfy process working directory. The output is a native Comfy VIDEO, so it can connect directly to nodes such as VACE Two Video Bridge Prep.

Direct two-video bridge

For editor/NLE use, prefer the VACE Two Video Bridge Prep and VACE Two Video Bridge Assemble nodes.

VACE Two Video Bridge Prep takes two native VIDEO inputs:

left_video: clip before the seam
right_video: clip after the seam
left_replace_frames: frames replaced from the end of the left video
right_replace_frames: frames replaced from the start of the right video
edge_blend_frames: outer bridge frames used as source anchors and final blend edges
fps: optional int/float force-input override; leave disconnected to derive and validate FPS from both videos
resize_strategy: passthrough keeps the current matching-input-size behavior; explicit_resize stretches both videos to width and height
width / height: optional int force-input targets used only when resize_strategy is explicit_resize
color_reference_frame: one-frame output for color matching; uses the last non-gray left control anchor, falling back to the left span/input frame

The visible bridge span is:

left_replace_frames + right_replace_frames

The node hides Wan's 4n + 1 generation length requirement by inserting masked padding at the seam. For example, a 32-frame visible bridge becomes a 33-frame Wan control window, and the assemble node trims it back to 32 frames.

VACE Two Video Bridge Assemble consumes the generated bridge frames after VACE decode and returns both:

bridge-only images/video
full joined images/video

Its edge_blend_easing setting controls the same edge-blend curve that the old crossfade node used when blending generated bridge edges back into source frames.

What this pack does not do

It does not bundle WanVideoNAG
It does not bundle ColorMatch
It does not bundle folder-manifest planning from the old workflow

If your workflow uses WanVideoNAG or ColorMatch, those still come from external node packs such as KJNodes.

Installation

Clone or copy this repo into your ComfyUI custom_nodes directory:

ComfyUI/custom_nodes/comfy-vace-automation

Then restart ComfyUI.

Dependencies

No extra pip packages are required beyond the normal ComfyUI runtime for the core nodes in this repo.

Notes:

torch is expected to come from the ComfyUI environment.
The nodes use ComfyUI internals such as nodes, comfy_execution, and comfy_api.latest.
Example workflows may still rely on external packs for quality extras such as WanVideoNAG and ColorMatch.

Repo layout

Status

This repo now tracks the newer in-memory VACE join flow and supersedes the old manifest/file-based automation pack that previously lived here.

Donations & Support

If this saves you time, you can support the work here: