Burve System Instructions

Small node, small job: it holds a set of pre-written system instructions for the Gemini generator and hands you the one you pick from a dropdown, as a string. That's it. No inputs beyond the dropdown, one instruction output.

Why bother, when you could just paste text into the generator's system_instructions field? Because these presets are meant to be packaged behavior. The whole point of a system instruction is that it steers the model's behavior no matter what the prompt says - so a curated, reusable set beats re-typing the same paragraph into every workflow. The pack's README and the JSON it ships make the intent clear: this is a library of instructions, and the dropdown is the picker.

Right now the library is deliberately small - two entries, None and "Face similarity request" - and the second one is the interesting one. It tells the model: if a reference face image is attached, incorporate that person's exact facial features and identity into the generated character, and prioritize that fidelity in all outputs. That's the pack's face-lock mechanism exposed as a preset, the same behavior Burve Character Planner emits as system instructions when you enable use_face_reference. It's the difference between "generate a woman with long hair" and "generate this specific woman."

How it works

The instruction text lives in a JSON file (system_instructions.json) that ships with the pack, and the node auto-checks GitHub for a newer version when it loads - same pattern as Burve Prompt Database. The instruction_name dropdown is built from whatever the file currently contains, so if the author adds presets in a future release, your dropdown grows without a manual update. The instruction output is the full text of the selected preset (an empty string for None), which wires directly into the generator's system_instructions input.

Inputs and outputs

• instruction_name - the dropdown; currently None or Face similarity request.
• instruction - the selected instruction text, as a STRING.

Notes

The auto-update is a network call at import time, and it's best-effort - if you're offline it just keeps the bundled file, no failure. Nothing about this node is hardwired to the Gemini nodes either; it emits plain text, so you could route it anywhere a string makes sense. But realistically, you'll use it as the system-instructions source for Burve Google Image Gen or the Vertex twin, ideally paired with a face reference image in reference_images. Install is just the pack itself - git clone https://github.com/Burve/ComfyUI_Burve_Tools.git into custom_nodes or grab it via ComfyUI Manager, then restart.