😊 Z2J Mood

Mood is the least-prompted dimension of an image and the one people feel most when it's missing. "Melancholy, subdued, with an undercurrent of unease" is doing a lot of emotional work that no individual tag can carry. Z2J Mood generates that layer deterministically - one of the most pleasant surprises in the ComfyUI-Zero2JSON pack, because it turns "what's this image about" into a reproducible knob instead of a feeling you have to put into words at 2am.

It sits in the style family of the pack. The generated phrase pairs a primary emotion with a secondary one, an intensity, and a descriptor - so you get structured, layered atmosphere rather than a single flat adjective. In the FLUX2-JSON workflow it feeds the mood/atmosphere slot and lets the model's LLM-style encoder actually chew on the emotional register, which is exactly the kind of descriptive block that structured prompting does well.

How it works

Standard position-as-seed core: (seed, prompt_index, slot) hashed with xxhash32, then a template and one word from each pool. The mood_default.json profile has 4 templates over four pools - emotion (24), secondary_emotion (12), intensity (12), descriptor (10) - about 138,000 combinations. That's a real emotional vocabulary: primary emotion, a complicating secondary, how hard to push it, and a textural descriptor.

The inputs that matter

• seed - world seed.
• prompt_index - position coordinate; farm this to walk the emotional space.
• profile - dropdown of mood_*.json files; add your own to profiles/ and restart.
• prefix / suffix - anchors around the mood phrase.
• mood_valence - a filter dropdown (positive / negative / neutral / tense / peaceful). Read the small print: in v1.0.0 as shipped this input is accepted but ignored by the generator. It doesn't constrain anything yet; check for a newer release.

Output is a single text string.

Install

One repo, all seventeen nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/MushroomFleet/ComfyUI-Zero2JSON
cd ComfyUI-Zero2JSON
pip install -r requirements.txt

Restart ComfyUI. The only dependency is xxhash>=3.0.0 - no models, no API, nothing heavy. Manager users: search "ComfyUI-Zero2JSON".

Where people get burned

Mood and lighting overlap more than people expect - a mood phrase like "foreboding" and a lighting phrase like "harsh key light from below" are telling the model the same story twice, and stacking them is usually fine but occasionally over-eggs it into camp. If a mood output feels overwrought, drop the lighting node's mood channel rather than the mood node. Also: mood words are weak if the scene doesn't support them - "serene" on a burning city reads as a prompt bug, not a subtle choice. Keep the generated mood and scene in the same emotional world.