Master Prompt Loader

If you've used this pack's main node for more than an afternoon, you know the drill: every captioning job wants a different system prompt, and typing a 200-word instruction block into a text box gets old fast. Master Prompt Loader is the pack's answer - a standalone dropdown that hands you a saved system prompt, ready to wire into anything.

It's the same system-prompt mechanism that's built into the big node, pulled out so you can use it on its own. The main Qwen node has a system_preset dropdown built in, and honestly, for a simple workflow that's the path of least resistance. Where this standalone version earns its keep: when you want the system prompt as a value you can inspect, branch on, or feed to a node that isn't the Qwen model at all.

How it works

The preset list is loaded from the pack's config JSON - specifically the _system_prompts section of system_prompts_user.json. The pack ships with 65 built-in presets (Tags, Simple Description, Detailed Analysis, JSON structure, and so on), and you can add your own by editing that file. The default system_prompts.json in the node folder gets overwritten on updates, so put your own additions in system_prompts_user.json in the pack folder or in ComfyUI/user/SimpleQwenVL_configs/ - the latter is created automatically and is the safer home.

• system_preset - the dropdown of saved prompts.
• system_prompt_opt - optional extra text appended to the chosen preset. Handy for a per-run tweak on top of a saved baseline.

Output: system_prompt, which is exactly preset + appended text. Wire it into the main node's system_prompt_override input and it takes precedence over whatever's in the dropdown there.

Install

It ships in the KLL535/ComfyUI_Simple_Qwen3-VL-gguf pack, so if you already installed that for the main Qwen node, you have this - it shows up in the same 🌐 SimpleQwenVL category. Fresh setup is the usual:

cd ComfyUI/custom_nodes
git clone https://github.com/KLL535/ComfyUI_Simple_Qwen3-VL-gguf

or search ComfyUI_Simple_Qwen3-VL-gguf in ComfyUI Manager, then restart and F5.

The practical pattern

The setup people actually run: one Master Prompt Loader feeding the main node's system override, a Style Selector feeding the user prompt, and a Remove Think node on the output. That's a reusable "caption this properly" block you can copy between workflows. The loader doesn't touch the model - no llama-cpp dependency, no VRAM, nothing to go wrong beyond a missing preset - so it's about as safe a node as this pack has.

One warning: if you edit system_prompts_user.json while ComfyUI is running, the dropdown won't always refresh until you restart the frontend. It's cached. Save your edits, F5, and if a new preset still doesn't show up, restart ComfyUI properly.