GALIAIS-Nodes AI Style Enhancer

You've got a prompt that describes the scene fine but renders flat - no lighting, no mood, no look. The AI Style Enhancer is the GALIAIS-Nodes answer: hand it your prompt plus a direction in plain words ("moody neon noir", "soft watercolor, morning light"), and it asks your language model to enhance the visual style while preserving the original tags. The "preserve the tags" part is the contract that keeps it from rewriting your whole prompt into something you don't recognize.

This is the LLM-in-the-graph enhancement pattern again, and it's one of the few places an API call genuinely earns its keep: style is exactly the thing people over-describe badly ("masterpiece, best quality" spam) and an LLM handles decently ("dramatic rim lighting, muted teal palette, shallow depth of field").

Inputs and outputs

Three inputs:

• AI服务商 - provider object from GaliaisNodesAIProvider. Live API call, so provider required.
• 提示词 (prompt) - your existing prompt, tags or prose, whichever your base model prefers.
• 风格方向 (style direction) - free text describing the look you want. This is the knob you actually tune. Vague in, vague out; "cinematic" gets you generic, "1980s VHS, film grain, warm halation" gets you something you can see.

Outputs: 风格增强提示词 (the enhanced prompt) and 原始响应JSON (raw response). Wire the first into your positive conditioning, keep the second for debugging.

The honest read

The main risk is the classic enhancer failure: the model adds detail you never asked for. A style direction can bleed into subject matter if the model gets chatty - "moody noir" becomes "detective with a cigarette" when you only wanted the lighting. That's why the "preserve the original tags" framing matters, and why you should eyeball the output rather than auto-feed it downstream. Also remember: every run is a metered API call, so this is a per-iteration cost, not a one-time setup.

It pairs naturally with this pack's tag-to-prose pipeline: run the AI Tag Analyzer, build your character sections, and use the Style Enhancer as a final pass when the composition is solid but the render feels flat. Skip it if your model is a pure tag-driven SDXL finetune where extra prose-style tags dilute - there, a sharper 画面风格 field in the scene node is free and local.

Install and gotchas

Same one-time pack install:

git clone https://github.com/GALIAIS/GALIAIS-Nodes.git ComfyUI/custom_nodes/GALIAIS-Nodes

Restart ComfyUI, or search "GALIAIS-Nodes" in ComfyUI Manager. No pip dependencies, no bundled models. Gotchas: configure the AI Provider first (Health Check is your friend), the pack is new with no community writeups, and if the style comes out wrong, spend effort on the 风格方向 text - that's the input that does the work.