NVIDIA/comfy_nv_video_prep

NVIDIA/comfy_nv_video_prep

Description: ComfyUI Node for easy video preparation making TTM use easy

Language: JavaScript

License: NOASSERTION

Stars: 3

Forks: 0

Open issues: 0

Created: 2026-03-02T10:02:13Z

Pushed: 2026-03-11T16:55:21Z

Default branch: main

Fork: no

Archived: no

README:

ComfyUI_VideoPrep

NVIDIA Video Preparation nodes for ComfyUI — a suite of interactive tools for cropping, mask editing, segmentation, keyframe animation, and multi-layer compositing, designed for video generation pre-production workflows.

Overview

ComfyUI_VideoPrep provides 5 custom nodes that keep the entire video preparation stage inside ComfyUI. Artists can interactively crop, segment, edit masks, and keyframe-animate layers without leaving the canvas, then pipe results directly into downstream generation nodes (Wan, LTX, etc.).

Nodes included:

| Node | Description | |------|-------------| | VideoPrep Interactive Crop | Draggable crop rectangle with live preview; simultaneous image + mask cropping | | VideoPrep Preview Bridge | Image/mask pass-through with premultiplied alpha and multiple output modes | | VideoPrep Preview Bridge (Advanced) | Built-in persistent mask editor with brush/erase and Cut/Keep/Extend modes | | VideoPrep TTM Interactive Editor | Click-to-segment with SAM2, polygon editing, multi-layer keyframe animation | | VideoPrep TTM Compose & Animate | Multi-layer compositor with up to 4 subjects, background pan/zoom/rotate |

Getting Started

# Clone the repository
git clone ssh://git@gitlab-master.nvidia.com:12051/awd/alt_comfyui_ttm_nodes.git

# Copy to ComfyUI custom_nodes
cp -r alt_comfyui_ttm_nodes /path/to/ComfyUI/custom_nodes/ComfyUI_VideoPrep

# Install dependencies
pip install -r /path/to/ComfyUI/custom_nodes/ComfyUI_VideoPrep/requirements.txt

# Restart ComfyUI — VideoPrep nodes will appear in the node menu

Requirements

• OS: Windows 10/11, Linux (Ubuntu 20.04+)
• Runtime: Python 3.10+, ComfyUI (latest stable)
• GPU: NVIDIA GPU with CUDA support
• VRAM (Minimum): 4 GB
• VRAM (Recommended): 8 GB+ (for SAM2 segmentation features)

Python Dependencies

• numpy
• opencv-python
• imageio, imageio-ffmpeg

Optional Dependencies

• [comfyui-sam2](https://github.com/neverbiasu/comfyui-sam2) — for click-to-segment in the Interactive Editor. Install via ComfyUI Manager. SAM2 models auto-download on first use.

Usage

Load one of the example workflows from the workflows/ directory:

• example_basic.json — Basic interactive editor usage
• example_compose_animate.json — Multi-layer composition and animation
• example_with_wan.json — Integration with Wan video generation pipeline

Node Details

VideoPrep Interactive Crop

• Click and drag to draw a crop region directly on the image
• Resize by dragging corners/edges; move by dragging inside
• Double-click to reset; grid snapping for latent alignment
• Inputs: image, mask (optional), crop_data, snap_to_grid
• Outputs: cropped_image, cropped_mask, x, y, width, height

VideoPrep Preview Bridge

• Handles both images and masks reliably across workflow runs
• Output modes: original, premultiplied, cutout_black, cutout_white
• Preview modes: cutout, overlay, mask_only
• Inputs: image, mask (optional), output_mode, preview_mode
• Outputs: image, mask, width, height

VideoPrep Preview Bridge (Advanced)

• Draw/erase masks directly on the image preview
• Mask stored inside the workflow file — persists across runs, saves, and restarts
• Multiple mask modes: Cut, Keep, Extend
• Auto-imports existing alpha channels from input images
• Inputs: image, mask (optional), output_mode, mask_combine, mask_data
• Outputs: image, mask, preview, width, height

VideoPrep TTM Compose & Animate

• Up to 4 animated subjects over an animated background
• Per-layer keyframe transforms: position, rotation, scale
• Background animation: pan, zoom, rotate with independent keyframe track
• Timeline scrubber for animation preview
• Inputs: background, up to 4× (subject_N, mask_N), editor_data, total_frames
• Outputs: frames, width, height

VideoPrep TTM Interactive Editor

• Click-to-segment using SAM2 (optional) for detailed mask creation
• Include/exclude points for precise segmentation control
• Manual polygon drawing fallback (if SAM2 is not installed)
• Multi-layer keyframe animation: position, rotation, scale, hue
• Animation timeline preview slider
• Inputs: image, editor_data, total_frames
• Outputs: frames, masks, width, height

Releases & Roadmap

• Releases / Changelog: [CHANGELOG.md](CHANGELOG.md)

Contribution Guidelines

This project is currently not accepting any contributions.

All source code is original NVIDIA-authored work. No third-party code is included.

• Code of Conduct: [CODE_OF_CONDUCT.md](CODE_OF_CONDUCT.md)
• Contributing policy: [CONTRIBUTING.md](CONTRIBUTING.md)

Governance & Maintainers

• Governance: [GOVERNANCE.md](GOVERNANCE.md)
• Maintainers: NVIDIA AWD Team — see [MAINTAINERS.md](MAINTAINERS.md)

Security

• Vulnerability disclosure: [SECURITY.md](SECURITY.md)
• Do not file public issues for security reports. Report to [psirt@nvidia.com](mailto:psirt@nvidia.com).

Support

• Level: Experimental
• How to get help: File an issue
• Details: [SUPPORT.md](SUPPORT.md)

License

This project is licensed under the Apache License 2.0 — see the [LICENSE](LICENSE) file for details.

Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.