Image Info Encryption Targets

This is the node that gives the Info-Prompt-Toolkit pack its reason to exist. ComfyUI's sharing culture runs on metadata: the workflow - prompt, model, seed, everything - is embedded in the output PNG, and dragging the image back reconstructs the whole generation. That's a feature right up until you want to post an image without the world seeing exactly how you prompted it. IPT-ImageInfoEncryptionTargets is your "what stays private" form: you tick the fields you don't want public, and the pack encrypts just those before the metadata is written, while keeping everything else in normal, A1111/Civitai-compatible infotext that still round-trips through Civitai and other tools.

This is the declaration node. It doesn't encrypt anything itself - it builds a list of targets. You then hand that list to IPT-SetImageInfoEncryptionTargets, which attaches it to your image_info, and the saver nodes (Image Saver, Referenced Image Saver, Video Saver) do the actual encrypting when they write the file.

How it works

Every input here is a boolean flag for one field of image_info: model, refiner, detailer, lora_stack, clip, vae, positive, negative, steps, sampler, scheduler, cfg, seed, and width_x_height (encrypts width and height together). Tick one and that field gets moved into the encrypted payload at save time instead of the public infotext. The last input, extra_keys, is a JSON list for image_info.extras - custom keys that aren't one of the fixed fields - and the node's frontend gives you a toggle editor for it rather than making you type JSON by hand.

The encryption itself is AES-256-GCM with a key you configure in ComfyUI Settings → Info-Prompt-Toolkit → Metadata Encryption. The default is a device-derived key, so it mostly "just works" on your own machine, but you can set a custom key and an optional fallback decryption key. Public fields stay plaintext and compatible with Civitai's metadata parser; the encrypted ones go into a private metadata area only the pack's reader nodes can restore.

The inputs that matter

• The 14 field toggles - positive and negative are the ones most people tick first; they're the most identifying content in the metadata.
• extra_keys - JSON rows for extras to encrypt (toggle editor in the UI).
• lora_stack and model - tick these if you don't want people seeing which LoRAs or checkpoint you used.

One output: encryption_targets, wired into IPT-SetImageInfoEncryptionTargets.

Installing it

Part of the IPT pack:

cd ComfyUI/custom_nodes
git clone https://github.com/kinorax/comfyui-info-prompt-toolkit.git
cd comfyui-info-prompt-toolkit
pip install -r requirements.txt

Restart ComfyUI (or use ComfyUI Manager, "ComfyUI-Info-Prompt-Toolkit"). The cryptography package in requirements is what powers the AES encryption, and the install pulls it in.

Where people get burned

The big one: your encrypted fields are only as recoverable as your key. The device-derived default is tied to this machine, so a file saved on one PC won't have its hidden fields restored on another - and if you set a custom key and lose it, those fields are gone forever. That's the design working as intended (that's the point of encryption), but back up the key if you care about recovering settings later. Second: encryption hides fields from other tools, but it doesn't make the image's pixels anonymous - the prompt may be locked away while the LoRA stack is still public if you didn't tick it. Third: the reader side needs the pack installed to unlock anything, so an encrypted image shared with someone who doesn't run IPT is an image with permanently missing metadata to them.