Ard Load Image

Here's the pack's party trick, and the reason its Save Image node exists: load a saved PNG, and Ard Load Image hands you not just the image and mask but the generation recipe - positive prompt, negative prompt, seed, steps, cfg, denoise, width, height. It's a "re-run this image" node: drop in a PNG, and everything you need to regenerate it (or continue from it) comes out as typed outputs ready to wire back into the sampler.

How it finds that info is worth knowing, because it explains when this works and when it silently doesn't. The node looks for two kinds of metadata embedded in the PNG:

1. ARD Save Image's own metadata - the pack writes a basic_metatags block into its saved PNGs, and the node parses that directly.
2. Standard ComfyUI metadata - failing that, it reads the prompt JSON that ComfyUI's own Save Image embeds, then scans the workflow graph inside it for the seed, steps, cfg, sampler, scheduler, denoise, checkpoint name, and prompt text fields, and pulls them out.

If neither source yields info (a random PNG, a screenshot, a JPEG - note the code only attempts extraction for .png files), the node prints a message and falls back to defaults: positive "error terminal", negative "blur, distortion", steps 7, cfg 3.0, seed 1234. In that state, only the image, mask, width, and height outputs are meaningful. That default prompt text is the pack's tell-tale - if you see "error terminal" in your positive prompt output, the image had no readable metadata and you're on defaults.

The controls: get_img_info (enabled/disabled - disabling makes the node act like a plain Load Image), seed_from (loaded image / random - pick random if you want a fresh seed every load), and terminal_info (print / dont print - dumps the extracted recipe to the console, which is genuinely handy for debugging).

Outputs, the full list: Image, Mask, pos_text, neg_text, seed, steps, cfg, denoise, Width, Height. Feed the text into an encoder (or this pack's Ard Dual Prompt), the numbers into the sampler or Control Box, and you've reconstructed the original run.

Honest verdict: this is the most genuinely useful idea in the pack - "give me the recipe from an image" is a real workflow (recreate someone's output, batch-similar generations, continue a composition). But the catch is the same as all metadata tricks: it only works with images that actually carry the metadata, which means files saved by this pack's Save Image or standard ComfyUI saves with embedded workflow. The KB's ComfyUI ecosystem material makes the same point - PNG metadata is how "workflow included" sharing works, and it only survives if the saving node writes it. If you load a PNG with no metadata, you get defaults and no warning beyond a console line, which is the #1 "why is my prompt 'error terminal'" confusion.

Install is the standard pack route - ComfyUI Manager → search ComfyUI-Ardenius, or:

cd ComfyUI/custom_nodes
git clone https://github.com/ArdeniusAI/ComfyUI-Ardenius

then restart ComfyUI. Pack-wide notes: civitai and moviepy install with the pack, and a console import error at startup is the Save Image node wanting comfyui_controlnet_aux - this node doesn't need that pack to work.