ComfyUI-Image-Safety-Gate

ComfyUI custom node combining three safety signals with boolean OR. Tuned for workflows that generate illustration / anime-style content.

Available in ComfyUI-Manager — search for ComfyUI-Image-Safety-Gate in the Custom Nodes Manager. See Install for details.

1. SmilingWolf wd-tagger v3 family (ONNX) — WaifuDiffusion tagger v3. Trained on Danbooru illustrations. Returns rating probabilities general / sensitive / questionable / explicit. We treat questionable + explicit as the unsafe signal. Five variants are selectable from the node (see WD tagger variants).
2. CompVis/stable-diffusion-safety-checker — CLIP-based concept similarity checker (same model the previous ComfyUI-safety-checker used). Sensitivity slider matches the old node.
3. OwenElliott/image-safety-classifier-s — SwiftFormer-based NSFL / NSFW / SFW classifier. We use only the NSFL dimension (gore / violence). The NSFW dimension over-triggers on illustrations, but the NSFL dimension empirically stays calm on illustrations and lights up on actual gore / violence.

Why this stack:

• WD tagger covers anime-style NSFW (sexual content) without illustration false positives.
• CLIP catches concept matches that WD might miss.
• image-safety-classifier-s NSFL covers gore / violence, which WD does not separate cleanly (Danbooru classifies blood / killing under "safe" rating).

OR keeps the policy conservative: block if any of the three flags the image.

Output

• IMAGE: input image, unmodified (no censoring is applied here)
• nsfw (BOOLEAN): True if any of WD tagger, CLIP, or NSFL flags it as unsafe

For multi-image batches, nsfw is True if any image in the batch is flagged.

Inputs

WD tagger variants

All variants are from SmilingWolf's v3 series. Larger models are slower / heavier but more accurate. All produce the same rating layout, so swapping variants does not require threshold retuning.

The selected variant is downloaded on first use into ComfyUI/models/safety_checker/wd-<variant>-tagger-v3/. Each variant is cached separately, so switching does not re-download.

Sensitivity / threshold reference

CLIP sensitivity:

• 0.0: least sensitive
• 0.5: explicit nudity threshold
• 1.0: most sensitive (catches lingerie-like content)

WD tagger threshold (sum of questionable + explicit):

• >= 0.50: high confidence NSFW
• 0.30 - 0.50: borderline / sensitive content
• < 0.30: likely SFW

NSFL threshold (image-safety-classifier-s NSFL dimension):

• >= 0.50: clear gore / dismemberment / corpse content (default)
• 0.15 - 0.50: light blood / wound content (lower catches more)
• < 0.15: essentially no violence signal — most illustration content sits here regardless of depicted scene

Install

Option 1: ComfyUI-Manager (recommended)

Available in the ComfyUI-Manager registry since Comfy-Org/ComfyUI-Manager#3007.

1. Open Manager → Custom Nodes Manager in ComfyUI.
2. Search for ComfyUI-Image-Safety-Gate.
3. Click Install, then restart ComfyUI.

Dependencies from requirements.txt are installed automatically.

Option 2: Manual (git clone)

cd ComfyUI/custom_nodes
git clone https://github.com/monkeykim111/ComfyUI-Image-Safety-Gate
pip install -r ComfyUI-Image-Safety-Gate/requirements.txt

Restart ComfyUI.

Model Files

On first use, the three models download into the standard safety_checker folder:

ComfyUI/models/safety_checker/wd-<variant>-tagger-v3/
├── model.onnx
└── selected_tags.csv

ComfyUI/models/safety_checker/stable-diffusion-safety-checker/
├── config.json
├── preprocessor_config.json
└── pytorch_model.bin

ComfyUI/models/safety_checker/image-safety-classifier-s/
├── config.json
└── model.safetensors

If files already exist (e.g., previously downloaded for ComfyUI-safety-checker or ComfyUI-image-safety-classifier), no network access is made.

Notes

• All models are cached in memory per process. WD variants are cached per variant key, so toggling between variants in a session does not reload.
• WD tagger runs via onnxruntime (CPU or GPU automatically depending on the installed provider). CLIP runs on CPU to match the original node's exact numerical behaviour. NSFL runs on GPU when CUDA is available.
• Per-image log line shows each signal's decision and the final OR result, so disagreement cases are visible for later threshold tuning.