🦾 Prompt with Selectors

If you're on an LLM-encoded model - FLUX, Chroma, Qwen, Klein, Anima, that family - and you keep typing the same "golden hour, over-the-shoulder" phrases into every prompt, this node exists to stop that. It's a structured prompt builder with two dropdown selectors that append clean, separated sections to your text instead of you retyping them and hoping the wording stays consistent.

How it works

The inputs are the main text field (multiline, where your real prompt goes) and two dropdowns: camera_angle and lighting. Each dropdown has a sentinel default - literally 📷 Select camera angle (optional) / 💡 Select lighting (optional) - plus the real options (nine angles like front view, top-down, low angle; five lighting modes from studio lighting to neon cyberpunk).

The behavior that matters: if you leave a dropdown on the sentinel, nothing gets appended. Your text passes through untouched. Pick a real value and the node appends it as a ### section on its own line:

a knight in a rain-soaked courtyard
### camera angle: low angle

### lighting: golden hour

That ### section style is the point. The community's structured-prompt findings (in our KB's prompt-engineering essay) keep landing on the same idea: LLM-encoded models read the prompt as a message, and they respond well to clean separation between blocks - JSON, headings, or a well-organized paragraph all do the same work. This node produces exactly that shape for camera and lighting, so on those models the structure reads naturally rather than being ignored.

The honest take

It's a convenience wrapper, not a capability. On SDXL or other CLIP-encoder models the ### headers are just prompt text - harmless, but they won't unlock anything and can read as odd tokens. And the dropdowns are fixed lists; there's no custom-value slot. If you need "3/4 angle from behind-left at dusk," type it yourself. Where this node earns its keep is batch consistency: twenty generations, same subject, flip between "golden hour" and "neon cyberpunk" between runs without touching the main prompt. That's a genuinely nice workflow when you're mood-boarding.

Installation

It ships in the Beyond Nodes pack - install once, get all of it:

• ComfyUI Manager - search "ComfyUI-Beyond_nodes" (or "Beyond Nodes"), install.
• Or manually:

  cd ComfyUI/custom_nodes
  git clone https://github.com/beyondprompting/ComfyUI-Beyond_nodes
  

  Restart afterwards. No requirements.txt, no model downloads, no extra dependencies - it's pure string handling on top of what ComfyUI already ships.

Common issues

• "My prompt suddenly has ### lines I didn't write." You left a dropdown on a real value instead of the sentinel. Go back to "Select … (optional)" to disable a section.
• "I changed the dropdown and nothing changed." That's the sentinel doing its job - the default is deliberately "no-op." Pick an actual option.
• Wire it where? The prompt output is a plain STRING, so it plugs into your usual CLIPTextEncode / TextEncodeFlux / whatever text encoder your model uses. No special connection needed.

Small pack, small footprint - there's no crowd-sourced workflow to copy, so the node is easy to test in isolation. Make a prompt, flip a dropdown, read the output. It does exactly what it says, and that's rare enough to appreciate.