Prompt Generator 🐑

PromptGenerator is the general-purpose prompt enhancer of the MieNodes pack. Where the Krea2/LTX2/Z-Image nodes are tuned for one model each, this one is the model-agnostic workhorse: give it a rough idea and an LLM connector, and it returns an enriched prompt. It's the one to reach for when you just want a better prompt and don't need model-specific shaping.

It lives in ComfyUI-MieNodes, MieMieeeee's utility pack. Same caveat as its siblings - it's a front-end that calls whatever LLM service you connect; it doesn't run a model itself.

What it's good for

Prompt writing is a real skill, and most people are okay-not-great at it. The prompt-engineering KB is a whole essay on the fact that 2026-era models want descriptive, well-structured natural language rather than tag soup - and that turning a vague idea into that is the actual work. This node hands that work to an LLM. You type "ethereal digital portrait," it comes back with the lighting, palette, texture, and composition detail filled in. It's also handy for translation and cleanup - feed it a prompt in another language or a messy one and let the model tidy it.

The pack's own README uses it exactly this way, generating a detailed artistic prompt for an ethereal portrait and displaying the result with a Show Anything node.

How it works

• llm_service_connector (required) - from a Set*LLMServiceConnector node. The LLM doing the enhancing.
• input_text (required, multiline) - your raw idea.
• mode (enum: simple or advanced, default advanced) - the important dial. simple does a lighter touch (cleanup, light enrichment, translation); advanced goes full detail-enrichment, turning a phrase into a paragraph. Use advanced when you want the model to do the heavy lifting, simple when you want your intent kept mostly intact.
• seed (required) - reproducibility and variation.

The single output is prompt (STRING). Send it to your text-encode, or preview with a ShowAnything node first.

Want to steer the style of enrichment beyond the two modes? The pack lets you register your own system prompt with AddCustomSystemPrompt, so you can bake your own house style into the enhancement instead of taking the default.

Installing it

Install the pack. ComfyUI Manager → search ComfyUI-MieNodes (ComfyUI_MieNodes) → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/MieMieeeee/ComfyUI-MieNodes

then restart. No model download for the pack. Nodes appear under the sheep-emoji 🐑 MieNodes menu.

The API key step

Needs an LLM, so needs an API key on the connector - api_token on the node, or mie_llm_keys.json (copy the .example in the plugin folder). ZhiPu and SiliconFlow have free-tier models to test with; grab a current free id from their site.

Common issues

Errors on run = connector, almost always. Missing/bad key or a config_key pointing at an empty entry. Confirm with CheckLLMServiceConnectivity first.

advanced can over-write your intent. The enrichment sometimes adds detail you didn't want or drifts from the core idea. If the result feels like a different picture than you asked for, drop to simple mode, lower the connector's temperature, or tighten your input_text.

It's model-agnostic, which means generic. If you're targeting a specific model, the tuned nodes (Krea2, LTX2, Z-Image) will fit that model's quirks better than this general one. Use this when you want flexibility, those when you want fit.