Parameter Parser 🪽

A raw metadata string is a wall of text. Parameter Parser 🪽 is the node that slices it into useful pieces. You feed it the raw metadata (from Load Image + Metadata, or pasted in), and it hands back eight outputs - a formatted summary plus separate, correctly-typed values for the model, sampler, seed, steps, CFG, and both prompts. If you've ever wanted to grab the seed out of a downloaded image so you can reproduce it, this is the node.

What comes out

• formatted - a human-readable multi-line summary: model, LoRAs, Seed | CFG | Steps | Sampler, resolution/denoise, and (optionally) the prompt.
• model - checkpoint/model name as a string.
• sampler - sampler name, with the scheduler appended when one's available (e.g. euler/ancestral).
• seed - as an actual INT, not a string. Feed it straight into a KSampler's seed.
• steps - INT.
• cfg - FLOAT.
• positive_prompt / negative_prompt - the two prompt strings.

That typed output is the whole point: seed is a number you can plug into a sampler, cfg is a float you can compare, not just print.

How it decides what it's looking at

The node sniffs the format before parsing: metadata starting with { is treated as ComfyUI's embedded prompt JSON; metadata containing a Steps: line is treated as the A1111/Civitai parameters dialect; anything else gets an attempt at the ComfyUI parser anyway. Empty input returns a harmless "No metadata" string and zeros - a much nicer failure mode than a crash. If it genuinely can't make sense of the text, formatted becomes "Could not parse metadata."

Two optional inputs give you control: show_prompt (whether the positive prompt appears in formatted) and max_prompt_length (truncation, 20–500, default 100). Prompts are long and ugly; you almost always want them truncated in the summary while the full text still lives in positive_prompt/negative_prompt.

The workflow it belongs to

The canonical chain is: Load Image + Metadata → metadata_raw → Parameter Parser → seed/sampler/steps/cfg into a Parameter Overlay (to bake a caption into the image), or seed into a KSampler to reproduce a look. It's also handy on its own: paste a parameters string from a Civitai page and immediately get the model name and seed as values you can wire anywhere.

Installing it

ComfyAngel installs once for all its nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/ThepExcel/ComfyAngel.git

Restart ComfyUI, or search "ComfyAngel" in ComfyUI Manager. Pillow is the only dependency - no model downloads.

The one thing that will always disappoint you: this parser can only extract what's actually in the file. Screenshots, JPEG re-encodes, and stripped uploads carry no metadata to parse - the text chunks are gone before you ever load them. Garbage in, "No metadata" out.