OpenShot-ComfyUI is a focused set of ComfyUI nodes built for OpenShot. It exists to bring useful modern AI models into real editing workflows with simpler, more reliable, OpenShot-friendly integrations than most demo-oriented community graphs.

Requirements

ComfyUI
PyTorch
ffmpeg / ffprobe
git
Python 3.10 or 3.11 recommended

Quick install

python -m pip install -r requirements.txt
python -m pip install --no-build-isolation git+https://github.com/facebookresearch/sam2.git

# Validate the install
python validate.py

Restart ComfyUI after install.

How this works

Wrap useful upstream models in simpler OpenShot-friendly nodes.
Prefer practical, reliable workflows over demo-style graphs.
Keep installs and first-use model downloads as simple as possible.

What this includes

OpenShotDownloadAndLoadSAM2Model
OpenShotSam2Segmentation (single-image)
OpenShotSam2VideoSegmentationAddPoints
OpenShotSam2VideoSegmentationChunked (meta-batch/chunk friendly)
OpenShotGroundingDinoDetect (text-prompted object detection -> mask + JSON)
OpenShotTransNetSceneDetect (direct TransNetV2 inference -> IN/OUT JSON ranges)
OpenShotDeepFilterNetDenoiseAudio (file-path based audio denoise -> FLAC path)
OpenShotLavaSRSpeechClarity (LavaSR speech runner -> FLAC path)

SAM2 nodes

These nodes are intended for practical OpenShot segmentation workflows, including image segmentation, promptable video segmentation, and chunk-friendly processing for longer clips.

OpenShotDownloadAndLoadSAM2Model downloads and loads supported SAM2 checkpoints
OpenShotSam2Segmentation handles single-image segmentation
OpenShotSam2VideoSegmentationAddPoints adds prompt points for video workflows
OpenShotSam2VideoSegmentationChunked is designed for chunked processing of longer videos

GroundingDINO and TransNetV2 are included alongside the SAM2 nodes to support object detection and scene boundary workflows that are useful in larger OpenShot pipelines.

Audio denoise node

OpenShotDeepFilterNetDenoiseAudio is intentionally minimal:

Input: source_audio_path, noise_reduction, keep_model_loaded
Output: a new FLAC file path
The node uses the upstream DeepFilterNet package and blends the enhanced result with the original signal using noise_reduction from 0.0 to 1.0
0.0 means "keep the original audio" and still emits a FLAC copy
1.0 means "full denoise"

The node accepts typical audio formats that ffmpeg can decode and always writes a new .flac file into ComfyUI's output folder under openshot_audio/.

Speech clarity node

OpenShotLavaSRSpeechClarity is intended for low-quality speech recordings:

Input: source_audio_path or audio, keep_model_loaded
Output: a new FLAC file path
Uses LavaSR v2 speech enhancement
Intended for speech/dialogue clarity, not general music restoration

LavaSR is installed through requirements.txt in the main Comfy environment. The first Clarity -> Speech run will still take longer because LavaSR downloads its model snapshot from Hugging Face on demand.

Troubleshooting

If python -m pip install -r requirements.txt fails on deepfilterlib with a Rust / Cargo error, you are most likely using Python 3.12+.

Simplest fix:

use a Python 3.10 or 3.11 Comfy environment

If you want to keep Python 3.12, install Rust first and rerun the install:

curl https://sh.rustup.rs -sSf | sh
source "$HOME/.cargo/env"
python -m pip install -r requirements.txt

Notes

OpenShotSam2VideoSegmentationChunked returns only the requested chunk range (bounded memory) instead of collecting whole-video masks.
For very long videos, pair chunked outputs with batch-safe downstream nodes (VHS meta-batch, staged processing, or on-disk intermediates).
torchaudio is listed explicitly because DeepFilterNet imports it internally and some environments do not pull it in automatically.
LavaSR preserves the original channel count by processing each channel independently before recombining the output.

Acknowledgements

OpenShot-ComfyUI builds on several excellent open-source projects and model releases. A big thank you to the maintainers, contributors, and researchers behind these repos and models. This node pack would not exist without their work.

Core upstream repos used by these nodes:

facebookresearch/sam2 for SAM 2 model code and checkpoints
IDEA-Research/GroundingDINO for GroundingDINO open-set object detection
soCzech/TransNetV2 and transnetv2-pytorch for scene / shot boundary detection
Rikorose/deepfilternet for DeepFilterNet audio denoising
ysharma3501/LavaSR for LavaSR speech enhancement
langtech-bsc/vocos via LavaSR for the underlying Vocos-based enhancement stack
kijai/ComfyUI-segment-anything-2 for integration ideas and surrounding ComfyUI ecosystem work

Please see the upstream repositories for full original licenses, credits, papers, and model details.

Licensed under the GNU General Public License v3.0 (GPLv3). See LICENSE.md for the full license text.