Nox Prompt Analyzer

Nox Prompt Analyzer is the simplest node in the Nox Prompter pack, and honestly the most underrated. You paste a prompt in, pick what you want back, and it gives you metrics: word count, a complexity score out of 10, a detected style, and a sentiment read. No AI, no API, no model - just deterministic heuristics that run in a few milliseconds. If you're tired of eyeballing whether your prompt has ballooned past what your checkpoint can hold together, this is a handy tripwire.

How it works

Pure Python string math, straight from the source. It splits the prompt into words for the count, then computes a complexity score from three factors: average word length (long words = denser), the ratio of unique to total words (more variety = more complex), and how many "artistic" terms like masterpiece, detailed, or artstation appear. The score is capped at 10. Style detection is a keyword dictionary - mentions of photography or dslr read as photorealistic, anime or manga as anime, cinematic or film as cinematic, and so on. Sentiment is a similar keyword pass.

analysis_type picks what you see: word_count, complexity, style_detection, or full_analysis (the default, which returns all of it plus a Short/Medium/Long length label). Feed it an empty prompt and you get ("Empty prompt", 0, 0.0).

Inputs and outputs that matter

Two required inputs, that's it: prompt and analysis_type. The outputs are the interesting part because they're typed, not just text: analysis_result is the human-readable STRING, but word_count is an actual INT and complexity_score a FLOAT. That means you can wire them into conditioning math or a batch logic node rather than just reading them - say, gate how many emphasis tags you add based on the returned word count.

Installing it

Pack install applies. ComfyUI Manager: search "Nox Prompter", install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/jamesk9526/NoxPrompter-ComfyUI-CustomNode

Restart, look for the NoxPrompter category. Zero dependencies beyond the standard library.

Gotchas

The main thing to know is what this node is not: it's not a real prompt-quality model, so treat its style and sentiment reads as rough guidance. "Detected style: photorealistic" means your prompt said words from the photoreal list, not that the node judged your writing. The complexity score also doesn't know which model family you're targeting - a long, nested prompt scores high here but may be exactly what an LLM-encoded model like Wan 2.2 wants, while a short tag list that a CLIP model would love scores low. Use it as a consistency check between batches, not as gospel. And don't expect presets - the Analyzer is one of the few nodes in the pack without preset_action/preset_name, which makes sense given it has no configuration worth saving.