ComfyUI-VideoSlice — ib video slicer

Custom ComfyUI node that reads .mov / .mp4 and outputs one frame as an IMAGE (plus sizes, slice index, and file frame counters).

Before you copy commands

• YOUR_REAL_GITHUB_USERNAME (or any YOUR_… URL in this doc) is a placeholder. Use your real login, e.g. https://github.com/octocat/ComfyUI-VideoSlice.git. Repository not found means the URL is wrong or the repo was not created on GitHub yet.
• /workspace/... paths are for RunPod (Linux pods) only. On your Mac, that folder does not exist—use Local Mac (no GitHub yet) or your real Comfy path.
• Do not paste whole blocks including lines that start with # unless you know they are comments. If the shell says command not found: #, you pasted a comment line as a command.
• cd: too many arguments: Usually a bad paste (extra words on the cd line) or a “smart” # character. Run cd on its own line with only the path, in quotes if the path has spaces.

---

Local Mac ComfyUI (no GitHub yet)

If the repo is not on GitHub yet, copy the folder into your local Comfy install:

# Example: adjust COMFY to where your ComfyUI lives
COMFY="$HOME/path/to/ComfyUI"
cp -R "/Users/ilyaduganov/Desktop/Comfy/Sliser__video/ComfyUI-VideoSlice" "$COMFY/custom_nodes/"

Then install deps with the same Python as Comfy:

cd "$COMFY/custom_nodes/ComfyUI-VideoSlice"
pip install -r requirements.txt

Restart ComfyUI.

---

Install from GitHub (any machine)

1. Clone into ComfyUI’s custom_nodes folder (name must stay ComfyUI-VideoSlice or Comfy will still load it if the folder contains __init__.py):

   cd /path/to/ComfyUI/custom_nodes
   git clone https://github.com/YOUR_REAL_GITHUB_USERNAME/ComfyUI-VideoSlice.git
   

2. Install opencv-python with the same Python / pip that runs ComfyUI:

   cd ComfyUI-VideoSlice
   pip install -r requirements.txt
   

   Or run the helper (tries common RunPod venv paths; see RunPod below):

   bash install_deps.sh
   

3. Restart ComfyUI (full process restart, not only refresh).

---

Publish this repo on GitHub

Fastest (GitHub CLI): install and log in once (brew install gh, then gh auth login), then from this repo:

bash create_github_repo.sh

That creates ComfyUI-VideoSlice under your GitHub user and pushes main.

---

From the folder that contains this README (skip git init if .git already exists):

git init
git add .
git commit -m "Initial commit: ib video slicer custom node"

On github.com/new: create a repository named ComfyUI-VideoSlice (empty repo, no template). Skip git init below if this folder is already a git repo.

git branch -M main
git remote add origin https://github.com/YOUR_REAL_GITHUB_USERNAME/ComfyUI-VideoSlice.git
git push -u origin main

If you see remote origin already exists, do not run remote add again—update the URL:

git remote set-url origin https://github.com/YOUR_REAL_GITHUB_USERNAME/ComfyUI-VideoSlice.git
git push -u origin main

Replace YOUR_REAL_GITHUB_USERNAME with your GitHub login in every clone/push URL.

---

RunPod ComfyUI

RunPod templates differ slightly; the idea is always: node code in custom_nodes/, deps in Comfy’s Python, videos in ComfyUI/input/ (upload or copy).

A. One-time setup on the pod

1. Open Terminal (or SSH) on the pod.

2. Find ComfyUI root (common cases):

   ls /workspace/ComfyUI/custom_nodes 2>/dev/null || ls /workspace/comfyui/custom_nodes 2>/dev/null
   

3. cd into custom_nodes on the pod (path varies by template; common: /workspace/ComfyUI/custom_nodes). Run cd alone on one line—no extra words after the path.

   cd /workspace/ComfyUI/custom_nodes
   git clone https://github.com/YOUR_REAL_GITHUB_USERNAME/ComfyUI-VideoSlice.git
   

   If that directory does not exist, find Comfy first: find /workspace -maxdepth 4 -type d -name custom_nodes 2>/dev/null

4. Install dependencies:

   cd ComfyUI-VideoSlice
   chmod +x install_deps.sh
   bash install_deps.sh
   

   If the script cannot find pip, locate Comfy’s venv and set COMFYUI_PIP:

   export COMFYUI_PIP=/workspace/ComfyUI/venv/bin/pip
   bash install_deps.sh
   

5. Restart the ComfyUI service / pod (per your template’s docs).

B. Persistence

• Ephemeral disks: reinstall or clone into a RunPod volume mounted at e.g. /workspace so custom_nodes survives stop/start.
• Put large videos on the same volume; place them under ComfyUI/input/ (or a subfolder) so they appear in the video dropdown.

C. Sharing with teammates

They use the same git clone URL into their pod’s custom_nodes, run install_deps.sh (or pip install -r requirements.txt), restart Comfy. No drag-and-drop install for the node itself—only media goes through the Comfy Input UI.

---

Comfy Registry (shows in ComfyUI-Manager)

There is no separate “merge into ComfyUI’s GitHub list” flow. Discoverability for installs like Manager → Install Custom Nodes comes from the Comfy Registry, which ComfyUI-Manager uses.

1. Create a publisher at registry.comfy.org (your Publisher ID is fixed; it appears after @ on your profile).

2. Create a Registry publishing API key under registry.comfy.org/nodes (not a normal GitHub token).

3. PublisherId is set in pyproject.toml (registry ftfftft).

4. Install comfy-cli and publish:

   comfy node publish
   

   GitHub Actions (auto-publish): Comfy-Org/publish-node-action. Repo secret name used here: REGISTER_ACTION_COMFYUI (Comfy Registry publishing key from registry.comfy.org/nodes). Workflow: .github/workflows/publish-to-comfy-registry.yml — on push to main when pyproject.toml changes, or workflow_dispatch. Bump version in pyproject.toml for each release.

   If git push rejects workflow files: your GitHub PAT needs the workflow scope (this is not the registry secret).

5. After publish, users can install via Manager / comfy node install ib-video-slicer (see registry for the exact id).

Docs: Publishing nodes, pyproject spec.

Node: ib video slicer (v1.1)

Category: video · Type IDs: IBVideoSlicer / VideoSliceFrame.

Video

• choose file to upload — one button; uploads .mp4/.mov into input/ (same API as image upload); refreshes the video list.
• video — file under input/ (relative path / subfolders allowed).

Inputs

| Input | Description | |--------|-------------| | frame_mode | increment / decrement / random over the slice list (per-video index state; random uses a new internal seed each run). | | lag_on | ~50%: random Δ ∈ {-3,-2,-1,1,2,3} in slice-index space, clamped; each time lag applies, Δ is never the same as the last applied Δ for that video. | | loop_on / loop_every_n_frames | Wrap window for increment/decrement. | | start_frame … skip_every_n_frames | Build slice list (from start through end, step skip_every_n). |

Outputs

| Output | Meaning | |--------|---------| | image, image_width, image_height | Usual IMAGE outputs. | | total_frames_count | INT — CAP_PROP_FRAME_COUNT for the file (total frames reported by the container/decoder). | | slice_index | INT — index in the slice list (0…n−1 after start/end/skip-every-n). |

Slice list

Range from start through end (-1 = last), step skip_every_n. Then frame_mode + optional lag.

Breaking: old widgets random_frame, random_seed, current_frame_index removed — use frame_mode and lag_on. Saved every_nth_frame is still read if present.

Migrating old workflows

Older graphs used every_nth_frame; the widget is now skip_every_n_frames (same meaning: 1 = all, 2 = every second, …). Re-open the node and set skip_every_n_frames to match the old value if you load an old JSON.

Supported formats

• .mp4, .mov