Quality-of-life custom nodes for ComfyUI: batch processing helpers, CSV-driven prompt iteration, file/folder loaders, string and JSON utilities, and integrations for ComfyDeploy, FAL, ElevenLabs, and Azure Blob Storage.

Originally bundled the now-abandoned Comfy_KepListStuff helpers.

workflow (1)

Install

Clone into ComfyUI/custom_nodes/:

cd ComfyUI/custom_nodes
git clone https://github.com/SozeInc/ComfyUI_Soze.git
pip install -r ComfyUI_Soze/requirements.txt

Or install via Comfy Registry / ComfyUI Manager.

Configuration

Some nodes need credentials in environment variables (or a .env file in the repo root):

| Variable | Used by | |---|---| | CD_API_KEY | ComfyDeploy nodes | | ELEVENLABS_API_KEY | ElevenLabs voice retriever | | SOZE_AZURE_STORAGE_CONNECTION_STRING | ComfyDeploy image parameter upload | | FAL_KEY (or [API] FAL_KEY in config.ini) | FAL nodes (e.g. Veo3.1) | | TENSORSCALE_API_KEY | TensorScale nodes. Keys are model-scoped, so you can instead set one per model: TENSORSCALE_API_KEY_MINIMAX_H3, _LTX2_5_FAST, _LTX_2_3_FAST, _COSMOS3_NANO, _HUNYUAN_IMAGE_3, _SENSENOVA_U1_5 (each falls back to TENSORSCALE_API_KEY). TENSORSCALE_BASE_URL overrides the host. |

The repo's .gitignore excludes .env and config.ini. Never commit credentials. If you suspect a key has leaked, rotate it.

Node groups

CSV — CSV Reader, CSV Reader X Checkpoint, CSV Reader X Lora, CSV Writer. The X Checkpoint/X Lora variants iterate index across rows × models, so a single counter walks every (row, checkpoint) pair.
Files / folders — Load Files From Folder, Load Files With Pattern, File Loader, Does File Exist, Save Image With Absolute Filename, Save Text File To Output, Append To Text File.
Images — Load Image, Load Images From Folder, Load Images From Folder X Lora, Load Image From URL, Load Image From Filepath, Image Batch Process Switch, Image Overlay, Empty Images, Variable Image Builder, Shrink Image, Pad Mask, Get Most Common Image Colors, Multi Image Batch, Soze Image Size With Maximum.
Video — Append To Video (ffmpeg concat with audio), Load Videos From Folder.
Strings — Multiline Concatenate, Multi Find And Replace, Special Character Replacer, Text Contains, String Functions, String Splitter, Empty String Replacement, Prompt Cache, Output Filename.
JSON — JSON Value Parser, JSON Path Extractor, JSON Array Iterator, JSON File Loader, JSON Formatter, JSON Get Array Count, Create Image Batch From JSON Array, Load Images From JSONArray.
Range / XY — Int/Float Range (Step), Int/Float Range (Num Steps), XY Any, XY Image.
Converters — Int/Float/Bool/String cross-conversion nodes, Boolean Inverter.
Loaders — Lora File Loader (load by absolute path), Checkpoint File Loader (load by absolute path).
Integrations — ComfyDeploy API queue / cache / download nodes, ElevenLabs voice retriever, FAL Veo3.1 ref-img-to-video.
TensorScale — synchronous world-model inference (no polling; each node POSTs once and streams the finished media back). TensorScale MiniMax H3 Video (fl2va) and ... Video Reference (ref2va) for video+audio, TensorScale LTX-2.5 Fast Video (video+synced audio, $0.04/video second), TensorScale LTX-2.3 Fast Video, TensorScale Cosmos3 Nano Text To Video / ... Video To Video, TensorScale Hunyuan Image 3 Fast, TensorScale SenseNova U1.5. Reference media is inlined as a base64 data URI by default; the whole request body is capped at 10 MiB, so use the *_url widgets or flip use_fal_upload on for anything video-sized. API docs

ffmpeg

Append To Video shells out to ffmpeg; it must be on PATH.

Contributing

Issues and PRs welcome. All respect to the original creators of any subsumed nodes — happy to upstream changes if you'd prefer them in your repo.

License

MIT — see LICENSE.