Style Applier

Style Applier is the cheat code of this pack, in the good way: no LLM, no Ollama, no waiting, no randomness. You give it a prompt and a style, and it returns the prompt with a fixed set of style keywords bolted on - plus a second output with just those keywords. Same idea as the generator's presets, but done with pure string logic, so it's instant and fully reproducible.

How it works

The node keeps a keyword catalog per style (in style_presets.py): primary descriptors, lighting, composition, texture, and technical terms. For cinematic that's the dramatic-lighting/color-palette/atmosphere vocabulary; for photorealistic it's the DSLR/85mm/camera-language set; video_wan gets video-relevant framing. Depending on the emphasis level and the include_technical toggle, it assembles a keyword string, then places it relative to your prompt based on position:

• suffix (default) - your prompt, <style keywords>
• prefix - <style keywords>, your prompt
• wrap - keywords on both sides

That's it. No model inference, no interpretation. If your prompt is empty, it just returns the keywords themselves.

Inputs and outputs

• prompt (required) - your base prompt.
• style - nine presets, the same list as the rest of the pack.
• position - suffix / prefix / wrap.
• emphasis - medium (default) / low / high. Note this is a keyword-detail level, not a numerical weight - don't expect it to scale emphasis the way a sampler weight would.
• include_technical (default on) - whether to include the camera/lens specs.

Two outputs:

• styled_prompt (STRING) - the finished prompt, straight into your text encoder.
• style_keywords (STRING) - just the keyword list, handy if you want to inspect it, save it, or wire it into another node without the full prompt.

Install

Same pack, same steps:

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

Under text/generation → Style Applier. Because it never calls Ollama, it works even if Ollama isn't installed at all - the one node in this pack you can add to a workflow without any LLM prerequisites.

When to reach for it (and when not to)

Reach for it when you want consistency - same style applied to a hundred prompts, identically, every run. The generator is a dice roll that occasionally gives you a beautiful surprise; Style Applier is a stamp. It's also the right tool when you already have a good prompt and just want the "cinematic" seasoning without a full re-roll.

The caveat is the same one that runs through this whole pack's era gap: keyword-append lands cleanly on SDXL-lineage models that parse a tag list, and much more weakly on LLM-encoded models (Flux, Z-Image, Anima), where a paragraph is an instruction and "8k, masterpiece" style tags are largely inert filler. On those models, position matters more than the vocabulary - a suffix is fine, but a wrap that fences your actual instruction with style keywords is the least likely to hurt. If you're on an anime SDXL model, this node is near-perfect: the tags are exactly the dialect those models speak.