Prompt Styler Original

If you've ever used the sai-cinematic, sai-anime, sai-neonpunk style names, this is where they come from. SDXLPromptStylerbyOriginal is the node in wolfden's pack that carries the original SDXL style templates - the ~100 presets Stability shipped through Clipdrop and that twri's sdxl_prompt_styler (the node this pack forks) made famous. This is the canonical set, not a themed spin-off. When people say "the SDXL prompt styler," these sai-* styles are usually what they mean.

Reach for it when you want the tried-and-true SDXL looks without reinventing them: cinematic, digital art, comic book, low poly, isometric, and the rest. They're battle-tested and they behave predictably.

How it works

Each style is a JSON entry with a {prompt} placeholder in its prompt field and a matching negative_prompt. Pick sai-enhance and a lighthouse at dawn becomes breathtaking a lighthouse at dawn . award-winning, professional, highly detailed. Pick sai-photographic and you get cinematic photo ... 35mm photograph, film, bokeh, professional, 4k. It's plain string substitution - no model, no API, no key. What makes it useful is that these particular templates were tuned against SDXL and genuinely nudge the output the way their names promise.

The negatives come along for free: each style has its own, and if you also type into text_negative, the two get merged rather than one replacing the other.

The inputs that matter

• text_positive - your subject, kept simple. The style adds the flourish.
• style - one of ~102 sai-* presets. This is the classic list.
• text_negative - optional, merged with the preset's negative.

log_prompt (No/Yes) echoes the final prompts to the console, which is the fastest way to learn what each sai- style is actually adding. Outputs are positive_prompt_text_g and negative_prompt_text_g - both strings. Send positive into the CLIP Text Encode feeding your positive conditioning, negative into the one feeding negative. On SDXL, the same string can go into both the G and L inputs of a CLIPTextEncodeSDXL.

Installing it

Via ComfyUI Manager: search SDXL Prompt Styler (wolfden), install, restart. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/wolfden/ComfyUi_PromptStylers

Restart ComfyUI. The node shows up under a Style Prompts submenu in the add-node menu. Nothing to download, no dependencies - it's just Python doing string work.

Where people get tripped up

These are SDXL styles, and that matters more than the "works with anything" convenience suggests. On SDXL and its lineage (Illustrious, Pony, NoobAI, WAI) they do what they say. Bolt them onto a 2026 LLM-encoded model like Z-Image or Flux 2 Klein and the quality-tag padding mostly evaporates, and the inherited negative does literally nothing at CFG 1 - where distilled Turbo/Lightning checkpoints run - because at CFG 1 the negative pass isn't even calculated. The node doesn't care what model you feed it, but the templates were written for CLIP-era SDXL.

If the node doesn't appear after install, restart fully and look under Style Prompts rather than the root menu. And if a style comes out heavier than you wanted, it's almost always the built-in negative - turn on log_prompt, read what it built, and trim it. The sai-* templates are just JSON in the pack folder, so you can open them, copy the format, and add your own. Since these are the reference styles, they're also the best ones to study if you want to understand the pattern before writing custom presets.