Cyberdelia Prompt Controls
A tiny pass-through that keeps your prompts in one place
- mode
- prompt
- use_vision
Cyberdelia Z-Engineer Input is the quiet half of this pack, and I'll be straight with you: the node does nothing. It has no logic beyond handing its two inputs straight back out. What it's for is keeping the things you actually type - your raw concept and the engineered-vs-raw toggle - in one tidy place instead of buried in the main Z-Engineer node's widgets.
What it is
It's a companion to Cyberdelia Z-Engineer (same pack, same category Cyberdelia/Prompt). You give it a mode toggle and a prompt field, and it routes them out as two connectable outputs. That's the entire route() function. The actual LLM call and CLIP encoding all still happen downstream in the main node - this one just exists to be a shared, visible control surface.
How to use it
The setup is two steps, and the second one is the part people forget:
- Add Cyberdelia Z-Engineer Input alongside your Z-Engineer node.
- In the Z-Engineer node, right-click the
modeandtextwidgets → Convert Widget to Input. - Wire the Input node's
modeoutput → Z-Engineermode, andprompt→text.
Now set the toggle and type your concept. The toggle label tells you what each side does: ✨ engineered (LLM) or → passthrough (raw). If you don't wire anything up, the Z-Engineer keeps working exactly as before from its own widgets - the companion is strictly additive, so existing workflows aren't disturbed.
Why you'd bother
For a single Z-Engineer node it's honestly optional. It earns its place when:
- You're running several Z-Engineer nodes off one concept - same raw prompt, different seeds or temperatures, and you don't want to retype it four times.
- You want the "raw idea" text visible at a glance rather than inside a node you're not looking at.
- You're driving the mode from other logic - an animated prompt or a control that flips engineered/passthrough mid-batch. The BOOLEAN output plugs into anything that wants a boolean.
Inputs and outputs
Inputs are just two: mode (BOOLEAN, defaults to true = engineered) and prompt (multiline STRING, your concept). Outputs mirror them exactly: mode (BOOLEAN) and prompt (STRING). There's no optional input, no hidden config - this is about as thin a node as the pack ships.
Installing and gotchas
Same install as the whole pack - it's one repository, so you get both nodes at once:
cd ComfyUI/custom_nodes
git clone https://github.com/cyberdeliaAI/comfyui-cyberdelia-z-engineer.git
pip install -r comfyui-cyberdelia-z-engineer/requirements.txt
or just search Cyberdelia in ComfyUI Manager and restart. Dependencies are a single requests line - nothing heavy, no model downloads.
Two gotchas worth naming:
- Forget Convert Widget to Input and the connection silently does nothing. The Z-Engineer's own widgets still hold the values, so your wired-in prompt just gets ignored. No error, no hint - it's a classic "why isn't this doing anything" moment.
- The default toggle means engineered mode. Default
mode = true, so with no LLM server running the main node falls back tofallback_input- you'll get your raw prompt back, which looks like the Input node "did nothing." Flip the toggle to passthrough (or start your LM Studio server) and the behavior snaps into focus.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| mode | BOOLEAN | true | — |
| prompt | STRING | — | |
| use_visionopt | BOOLEAN | false | — |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| mode | BOOLEAN | — |
| prompt | STRING | — |
| use_vision | BOOLEAN | — |