Extensions/Kennmarq.com: AI watermark and C2PA save nodes
ComfyUI Extension

Kennmarq.com: AI watermark and C2PA save nodes

Drop-in Save Image, Save Audio and Save Video nodes that mark AI output as it's saved: an invisible watermark and signed, time-stamped C2PA Content Credentials, the AI labels the EU AI Act (Art. 50) and California ask for. Images are marked on your machine; audio and video in Kennmarq.com's cloud. Example workflows and a local safety check included.

By Kennmarq-com·Created 23 days ago·Updated 12 days ago· 0
Kennmarq-com/comfyui-kennmarq
Nodes4
On cloudLocal install
CategoryKennmarq.com
Stars0
Updated12 days ago
Readme

Kennmarq.com for ComfyUI

Mark AI pictures, songs and clips for the EU AI Act as you save them. Kennmarq.com Mark & Save Image takes the place of Save Image and gives every picture an invisible TrustMark watermark and signed, time-stamped C2PA Content Credentials that say it was made with AI. They're made on your machine: only the manifest (about 13 KB, never pixels) goes to Kennmarq.com to be signed. Kennmarq.com helps you meet the Act's marking duty (Art. 50(2)); you stay responsible for your system.

Set up

  1. Install from ComfyUI-Manager (search "Kennmarq.com"), with comfy node install comfyui-kennmarq, or clone it into ComfyUI/custom_nodes and install its requirements in ComfyUI's Python: git clone https://github.com/Kennmarq-com/comfyui-kennmarq ComfyUI/custom_nodes/comfyui-kennmarq, then pip install -r ComfyUI/custom_nodes/comfyui-kennmarq/requirements.txt.
  2. Get a key at kennmarq.com/signup, free for 1,000 files a month with every feature. Start ComfyUI with it in KENNMARQ_API_KEY. The key is never a widget: a widget's value is saved in the workflow, and the workflow goes into every PNG you share.

Without a key, Mark & Save Image still writes the watermark and the IPTC AI label, and its notes say that signed Content Credentials need a key.

Example workflows

Two free workflows come with the nodes. In ComfyUI, open Workflow → Browse Templates → comfyui-kennmarq:

  • Text to image (example_workflows/kennmarq-text-to-image.json): Stable Diffusion 1.5, saved through Mark & Save Image.
  • Image to image (example_workflows/kennmarq-image-to-image.json): a picture from Load Image, repainted; source: auto marks it as composite, made from another picture.

Both load v1-5-pruned-emaonly-fp16.safetensors, and ComfyUI offers to download it when it's missing. To mark your own workflow, put Mark & Save Image in place of its Save Image node (and Mark & Save Audio or Video in place of Save Audio or Save Video).

The nodes

  • Kennmarq.com Mark & Save Image: PNG, JPEG or WebP.
    • source: auto reads the graph: a Load Image node means the picture was made from another one (composite), otherwise created.
    • keep_workflow: keeps ComfyUI's prompt and workflow in a PNG, as Save Image does, inside the signed file. They reveal your prompt; turn it off to leave them out.
    • safety_check: ai-act or safe-only runs the check below before saving; a blocked picture isn't written.
    • model, app_name: recorded in the Content Credentials.
  • Kennmarq.com Mark & Save Audio: an AUDIO saved as MP3, M4A or WAV, with Kennmarq.com's inaudible AudioSeal mark and signed Content Credentials. Five marks per started minute.
  • Kennmarq.com Mark & Save Video: frames (an IMAGE batch), their frame rate and an optional AUDIO, saved as an MP4 with Video Seal in every frame and AudioSeal in the sound. Its safety_check reads 12 frames (two marks each); a blocked frame leaves the clip unsaved. Five marks per started second. visible_label (or deepfake) draws the EU's "AI GENERATED" label on every frame, in English, Korean or Chinese, in the corner label_position names. Songs and clips are marked in Kennmarq.com's cloud, so these two nodes need KENNMARQ_API_KEY.
  • Kennmarq.com Safety Check: screens pictures for sexual content with someone who appears to be a child, and with a realistic face (the AI Act's Art. 5 bans), on your machine. Nothing is sent and nothing counted. It stops the run on a block, or passes its answers to Mark & Save Image, which then leaves blocked pictures unwritten. A risk signal with published error rates, not a guarantee. Its models download once (about 2.3 GB), and it needs onnxruntime.

The watermark's model downloads once (65 MB) to ~/.cache/kennmarq/models (or KENNMARQ_MODELS_DIR). On Windows, where the C2PA library has no build yet, the node makes the watermark on your machine and the Content Credentials in the cloud.

MIT-licensed.