Model Data (Mango)

Every checkpoint has an opinion about how it wants to be run. The Pony family wants its score tags and a CFG around 5–9; anime models want clip skip 2; some realism models basically want CFG near 1. Remembering all that per model is a fool's errand, which is what Model Data (Mango) is for: give it a checkpoint name and it looks up a bundled JSON file of per-model usage tips, then returns the recommended settings as a text snippet you can display or wire onward.

It's a reference card, not an executor. The node doesn't apply anything - it just fetches and formats. Think of it as the "don't forget to set clip skip 2 for this one" nudge built into your graph.

How it works

The node takes a STRING input - the checkpoint name, which typically comes from the ckpt_name output of Loader (Mango) or from a sampler's checkpoint field. It loads a model_data.json file that ships inside the pack and looks up your entry: first by the full key, then by the bare filename, then case-insensitively. If it finds a match, it returns the entry formatted as pretty-printed JSON. If not, it returns an error string that lists the available keys so you can see what's in the database.

What the database actually contains

The shipped model_data.json covers a handful of well-known community checkpoints - things like Pony Diffusion, various anime mixes, and realism models - and for each one records the author's recommended clip_skip, steps, cfg, sampler_name, scheduler, sample positive/negative prompts, and resolution. For instance, the Pony entry suggests clip skip 2, 25–30 steps, CFG 5–9, Euler a, and the classic score_9, score_8_up, score_7_up positive tags. It's a starting point someone typed by hand, so treat it as opinionated guidance, not gospel.

Inputs and outputs

There's exactly one input - ckpt_name (STRING, labeled "Checkpoint name") - and exactly one output: a STRING containing either the matched settings as JSON or a helpful error message. Because it's just a STRING, you can wire it to a Show Text node, save it with a text-saver, or feed it anywhere else text flows. It's the rare node where "no output if the model isn't in the DB" is actually fine, because the error text is genuinely useful.

Installing it

Model Data (Mango) comes with the Mango Node Pack:

cd ComfyUI/custom_nodes
git clone https://github.com/mang01010/MangoNodePack

Restart ComfyUI, or install "Mango Node Pack" through ComfyUI Manager. No extra dependencies beyond the README's pip install safetensors line.

Gotchas

The honest limitation: the bundled database only covers the models the author decided to include. If your checkpoint isn't in there, the node doesn't fetch anything from the internet - it just tells you the key is missing and lists what is available. If you want to extend it, you'd edit model_data.json inside the pack folder to add your own entries, which is a reasonable DIY for anyone comfortable with JSON. For a beginner, this node is a nice "what should I try first" reminder rather than a must-have; you can get the same info from any model card. But as a self-contained nudge in your workflow, it does the job.