System Prompt (Lumina 2)

Lumina 2 is one of those models that doesn't just take a prompt - it expects a system prompt prepended to it, a whole paragraph telling the model what kind of image-text alignment to use. If you've ever copied that paragraph from a workflow, noticed you got it subtly wrong, and then wondered why adherence changed, this node is for you: it stores the official system prompt strings and hands you the right one from a dropdown.

How it works

One input: system, a combo box of the Lumina 2 system prompt options (the default is superior). One output: system_prompt, the full text of the chosen prompt as a string.

The two headline options are the ones Lumina 2 actually documents: Superior ("generate superior images with the superior degree of image-text alignment...") and Alignment ("generate high-quality images with the highest degree of image-text alignment..."). The pack also carries the anime-flavored variants and a negative system prompt (the one that instructs the model to generate low-quality images, used for the negative side of conditioning), so you have the whole family in one dropdown instead of scattered across text nodes.

Why a dropdown beats a copied paragraph

Because the system prompt is long, exact, and identical every time. Memorizing it is a waste; pasting it from a tutorial is a vector for drift. This node guarantees the exact official string, and it slots naturally into the pack's Lumina 2 workflow: the pack's "Prompt Text (Lumina 2)" node combines a system prompt and your user prompt with the <Prompt Start> marker between them, and the example Lumina 2 workflow wires this node straight in. Select the prompt once, and both your positive and negative conditioning paths get the right system text.

Installing it

Part of the Sage Utils pack - ComfyUI Manager (search Sage Utils) or:

cd ComfyUI/custom_nodes
git clone https://github.com/arcum42/ComfyUI_SageUtils
cd ComfyUI_SageUtils
pip install -r requirements.txt

Restart ComfyUI. Only pip dependency is dynamicprompts; no model downloads - though the node is only useful if you're actually running a Lumina 2 checkpoint.

The thing to remember

The output is just a string - it won't do anything by itself. It needs to be combined with your prompt (via the pack's Prompt Text node or however you normally assemble Lumina 2 prompts) and then encoded. And note the "negative" option isn't a mistake: on the negative conditioning path, using the negative system prompt is what keeps your negatives working as intended with this model. Pick per conditioning path, not per workflow.