🗂️ Prompt Styles

If you've ever used the "style" dropdown in Fooocus or A1111 - the one that quietly wraps your prompt in a pile of quality tags and an aesthetic direction - this is that, as a ComfyUI node. Pick from a big library of pre-written style cards, stack up to six of them, and the node splices their text into your positive and negative prompts. It outputs plain strings, so it sits before your text encoder, not after. It's a shortcut for people who don't want to memorize the same "masterpiece, best quality, cinematic lighting…" boilerplate for the hundredth time.

How it works

A style card is just a saved prompt fragment with a placeholder for your subject - "cinematic still of {prompt}, dramatic lighting, 35mm" plus a matching negative. The node takes your bare positive and negative, finds the styles you selected, and merges the fragments in. Because the output is a STRING (not conditioning), you wire the results into a CLIP Text Encode as usual. Nothing magic is happening - it's prompt templating - but a good style library saves real time and helps beginners get "why does everyone else's output look so much more polished than mine" without learning the whole tag vocabulary first. The cards live in a CSV you can extend (the pack ships styles.csv and lets you add your own via my_styles.csv).

The inputs and outputs that matter

• positive and negative - your base prompt text. The subject you actually want; the styles wrap around it.
• style through style6 - six dropdowns, each a full library of 200-plus cards (you'll spot the Kamph_* set, base-prompt presets for 1.5 and XL, and many more). Stack several to layer effects. Cards are tagged for whether they contribute to the positive, the negative, or both.
• translate - a language dropdown; the node can translate your prompt into English first, which matters because the models are trained on English tags. Handy if you write prompts in another language.
• seed - drives any randomization in dynamic-prompt-style cards, so you can vary or lock results.

Outputs are the assembled positive and negative strings.

Installing it

ComfyUI Manager: search SDVN_Comfy_node, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/StableDiffusionVN/SDVN_Comfy_node

Then pip install -r custom_nodes/SDVN_Comfy_node/requirements.txt from your ComfyUI root, and restart. To customize the library, copy my_styles.csv.example to my_styles.csv and edit.

Where people get burned

Style cards are model-specific. A lot of these presets are written for SD 1.5 or SDXL and lean hard on booru-style quality tags (masterpiece, best quality). On Flux or other natural-language models, that tag soup does little or actively hurts - those models want sentences, not tag lists. Match the card to your model, and don't assume an XL preset helps on Flux.

Over-stacking. Six styles at once often fight each other and drown your actual subject. One or two focused cards usually beat a full stack.

Forgetting it outputs strings, not conditioning. The output goes into a text encode node, not straight into the sampler. If you're looking for a CONDITIONING output here, it doesn't exist - that's by design.