APNext Science

Want a black hole, a Bravais lattice, or a CT-scan aesthetic in your image but you don't know the right words to summon it? APNext Science is a dropdown that does the vocabulary for you. Pick from astronomy, mathematics, the periodic table, or medical imaging, and it appends those terms to your prompt. It's one node in a family - Scene, Time, Vehicle, Stuff all work identically - and the community knows the pack as the SDXL Auto Prompter.

This is the niche one. Most people reach for Scene or Stuff; Science is for when you specifically want that technical, clinical, or cosmic flavor and want to brainstorm terms you'd never think to type.

How it works

It's a menu, not a model. No API key, no download, nothing calls out to the network. The node holds curated lists baked into the pack - 92 mathematics terms, 33 astronomy, 121 chemical elements, 23 medical imaging modalities - and each dropdown lets you choose None, a specific term, Random, or Multiple Random. On run, it collects whatever each dropdown resolved to, joins them with your separator, and appends the lot to your base prompt. The rolls are seeded, so the same seed reproduces the same result. That's it: string assembly from menus.

The inputs and outputs that matter

• prompt (required, multiline) - your base text; picks get appended to it.
• separator (default ,) - the joiner between terms.
• The four category dropdowns - astronomy, mathematics, elements, medical - each None, a specific pick, Random, or Multiple Random.
• seed - fixes the random rolls for reproducibility.
• attributes (boolean, default off) - pulls in extra descriptive words when the pack's data attaches them to a term.
• string (optional) - extra text to merge in.

Outputs are two STRINGs: prompt (the assembled text) and random (the version with random picks resolved, so you can capture what it rolled). Feed either into a CLIP Text Encode.

How to install it

ComfyUI Manager is the simple route - search comfyui_dagthomas or "SDXL Auto Prompter", install, restart. Or manually:

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

Restart afterward. The requirements.txt is heavy (transformers, decord, provider SDKs) because the same pack ships a big suite of LLM and vision nodes - but Science needs none of it. It's a self-contained local text node.

Common issues & troubleshooting

The catch here isn't the node, it's your model's grasp of jargon. A term like "Apollonian gasket" or "Bravais lattice" only helps if the checkpoint actually learned that concept, and most general-purpose models learned the visually famous ones - black holes, galaxies, x-rays - far better than the obscure math. Start with the recognizable picks; treat the deep-catalog entries as experiments.

The bigger, structural gotcha is the pack's own nickname. SDXL Auto Prompter means this thing emits comma-separated tags, which is what CLIP-encoded models (SDXL, Illustrious, Pony) thrive on. On a 2026 LLM-encoded model - Flux, Z-Image, Anima - that comma soup fights the encoder, which reads a prompt as an instruction and starts drifting past ~75-100 tokens. So Multiple Random across every category is a great brainstorming toy for SDXL and a subject-burying mess on Flux. Use one Random category as a garnish there, not the whole dropdown wall.

And keep the pack current. It's large and under active development, so a stale checkout is the likeliest source of a load error - a git pull or Manager update has historically been the fix.