Prompt Styler Filter

Think of this as Instagram filters, except applied at generation time through the prompt instead of afterward in an editor. The dropdown is a short, practical list of photographic treatments: BlackAndWhite, Contrast, Cool, HDR, Monochrome, Saturated, Sepia, Vignette, Vintage, Warm. Pick one and the node appends the color and tone language that biases SDXL toward that look.

It's the most utilitarian node in wolfden's SDXL Prompt Styler pack - no fantasy worlds or dreamscapes, just grade-and-tone. Handy when you want a batch of images to share a consistent mood without hand-typing "warm tones, golden cast, soft contrast" every time.

How it works

Each filter is a JSON template with a {prompt} placeholder in its prompt field and a negative_prompt. Your positive text drops into the placeholder, the filter wording wraps around it, and the node returns the finished pair. Blank negative? You get the template's. Typed negative? Yours merges with it. No model, no GPU work, no download - the node writes text, and the checkpoint does the rendering.

One honest caveat: prompting for a "filter" biases the generation, it doesn't apply a deterministic LUT. Sepia will pull the whole image toward warm brown, but if you need an exact, repeatable grade, doing it as a post step (curves, a color-match node) gives you tighter control. This node is the quick, in-prompt version - great for exploration, looser than real color work.

The inputs and outputs that matter

• style - the filter. The reason the node exists; ten treatments to choose from.
• text_positive - your actual scene. The filter only changes how it's toned.
• text_negative - optional; blank uses the template's default negative, or add your own.

log_prompt (No/Yes) prints the assembled prompt so you can confirm, say, that Monochrome didn't quietly conflict with a color you named in your subject.

Outputs are positive_prompt_text_g and negative_prompt_text_g - two strings, wired into your positive and negative CLIP Text Encode nodes.

Installing it

Via ComfyUI Manager: search SDXL Prompt Styler (wolfden), install, restart. Manually:

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

Restart ComfyUI afterward. No dependencies, no model downloads. The styler nodes land under a Style Prompts submenu.

Common issues

Color contradictions. If your positive says "red dress" and you apply BlackAndWhite or Monochrome, the model has to reconcile two conflicting instructions and the result is a coin flip. Drop the specific colors from your subject when you're going grayscale.

none does nothing. The first entry, No option, is a deliberate passthrough. If the filter isn't landing, check you're not on it.

Weaker than a real grade. As above - this is a prompt bias, not a guaranteed tone curve. For precision, grade in post; for speed and vibe, this is fine.

SDXL-era tool. The templates were written for SDXL and rely on tag-style prompting plus a working negative. On distilled or LLM-encoded models (Flux, Z-Image) the negative is inert at CFG 1 and the extra words read as literal instruction, so the filter comes through faintly. Keep it on SDXL checkpoints.