Batch-process a folder of images or videos through any ComfyUI workflow. Captioning, upscaling, style transfer, video analysis, frame extraction — whatever your workflow does to a single file, these nodes make it do that to an entire folder. Click Queue Prompt once and walk away. Resume interrupted batches instantly with zero wasted GPU time.

Files load directly from disk. Nothing is copied into ComfyUI's input or cache directory.

Nodes

All nodes appear under 📂 BatchFolder in the node browser.

Load Image From Folder (Batch)

Loads one image at a time from a folder. In sequential mode the counter auto-advances each execution.

Outputs: IMAGE, folder_path, filename_stem, filename_full, current_index, total_images

Supported formats: PNG, JPG, JPEG, WebP, BMP, TIFF, GIF

Load Video From Folder (Batch)

Loads one video at a time from a folder, extracting frames as an IMAGE batch tensor.

Frame modes: first_frame, last_frame, all_frames, evenly_spaced (default 16 frames), frame_range

Outputs: FRAMES (IMAGE batch), folder_path, filename_stem, filename_full, current_index, total_videos, frame_count, fps, duration_seconds

Supported formats: MP4, AVI, MOV, MKV, WebM, FLV, WMV, M4V

Requires: opencv-python (pip install opencv-python)

Save Text File

Saves text to a file matching the source filename.

Queue Next

The loop engine. Place at the end of any workflow. After each execution it re-queues the workflow. When the loader runs out of files, execution stops cleanly. Has both image and text inputs — wire whichever is the last output of your chain.

Installation

Via Git Clone

cd ComfyUI/custom_nodes
git clone https://github.com/hodgemann/ComfyUI-BatchFolderTools.git
pip install opencv-python  # if not already installed

Restart ComfyUI.

Via ComfyUI Manager

Search for Batch Folder Tools in Install Custom Nodes, install, and restart.

Via Comfy Registry

pip install comfy-cli
comfy node install batch-folder-tools

Included Example Workflows

| File | What it does | |------|-------------| | batch_caption_qwenvl_mod.json | Image captioning — loads images, generates detailed captions via QwenVL-Mod, saves .txt files | | batch_video_caption_qwenvl_mod.json | Video captioning — loads videos, samples 16 evenly-spaced frames, generates captions, saves .txt files |

Both require ComfyUI-QwenVL-Mod. Pre-wired and ready to use — paste your folder path and click Queue Prompt.

Usage

Paste the full path to your folder into folder_path.
Set mode to sequential.
Click Queue Prompt once.
Walk away. Console shows progress. When done: ✅ DONE.
To stop mid-run, press Escape or click Cancel.

Resuming an Interrupted Batch

If the run gets interrupted (crash, power loss, you stopped it manually), set skip_captioned to true on the loader node and click Queue Prompt. The loader checks for an existing .txt, .caption, or .cap file next to each image/video — any file that already has a caption is skipped entirely with zero GPU cost. It picks up right where it left off.

Supported Path Formats

| Platform | Example | |----------|---------| | WSL | /mnt/c/Users/me/images or /mnt/d/datasets | | Linux | /home/user/images or /data/training_set | | macOS | /Users/me/images or /Volumes/External/photos | | Windows | C:\Users\me\images or D:\datasets | | Network shares | Any mounted path readable by the OS |

Tips

keep_model_loaded: true on VLM nodes — loads once, stays in VRAM.
max_tokens: 2048 for detailed descriptions.
temperature: 0.3 for consistent output; 0.6+ for variety.
skip_captioned: true to resume interrupted batches.
Filenames don't need to be sequential.

Changelog

v0.8.0

Added skip_captioned to both loaders — skips files that already have a caption, with zero GPU cost. Resume interrupted batches instantly.
Added fps and duration_seconds outputs to video loader.
Rewrote captioning prompts to only describe what is present (never mentions absences).
Removed audio inference from video prompt (QwenVL is vision-only).
Fixed duplicate captions by randomizing VLM seed on each re-queue.
Added GitHub Actions workflow for Comfy Registry auto-publishing.
Fixed pyproject.toml for Comfy Registry spec compliance.

v0.7.x

Added video loader with 5 frame modes. Separated Queue Next into its own node.

v0.6.0

Renamed to ComfyUI-BatchFolderTools.

v0.5.x

Self-re-queuing via hidden PROMPT input. Cache-busting.

v0.1.0–v0.4.x

Initial development.

License

MIT