Prompt Processing

Prompt Processing is the node that decides how much your style prompt matters, how much your negative style matters, and how strongly the conditioning is scaled - the tuning layer that sits on top of the five prompt boxes. If your Searge images ignore your style prompt, this is where you fix it.

It's part of SeargeSDXL, the base-and-refiner SDXL workflow pack Searge released when SDXL 1.0 dropped in 2023. The node is in Searge/_deprecated_/UI/Inputs - v4.x moved to a data-stream design - but the knobs it exposes are about prompt weighting, which is a timeless topic.

How it works

Searge splits a prompt into main / secondary / style / negative / negative-style parts (in the Prompts node). This node controls how those parts get blended into the actual conditioning fed to SDXL. style_prompt_power and negative_style_power are the balance knobs - how loud the style side is relative to the subject side. The _conditioning_scale values scale the overall conditioning strength for the base and refiner separately, which is a subtler lever most people leave alone.

Like the other input-chain nodes, it doesn't generate anything - it adds its settings to the PARAMETER_INPUTS bundle passing through, and the values get read back out downstream.

The inputs that matter

• style_prompt_power (FLOAT, default 0.33) - how much your style prompt influences the result. Turn this up if your aesthetic isn't landing.
• negative_style_power (FLOAT, default 0.67) - the same for the style-side negative.
• base_conditioning_scale / refiner_conditioning_scale (both FLOAT, default 2) - overall conditioning strength for each stage. Leave at default unless you know why you're changing it.
• style_template (enum: none, from preprocessor, test) - a hook for style templates; none is normal.
• inputs in/out (PARAMETER_INPUTS) - the chain wire.

How to install it

ComfyUI Manager: search SeargeSDXL, install, restart. Manual: cd ComfyUI/custom_nodes && git clone https://github.com/SeargeDP/SeargeSDXL.git, then restart. If you install manually, run python -m pip install opencv-python in ComfyUI's Python environment first - the README requires OpenCV or the nodes won't register. Generation needs the pack's models (SDXL base + refiner, fp16-fix VAE); the README lists them all.

Common issues

The number-one thing people hit: "my style prompt does nothing." Two culprits - style_prompt_power too low (bump it up), or the prompt style you selected in the Misc Parameters node doesn't use the style prompt at all (a few of the eleven styles deliberately ignore it). Check both.

Don't over-crank base_conditioning_scale chasing "stronger" prompt adherence - past a point it fries the image rather than sharpening the concept. Small moves. And the standing Searge rule applies: pair the latest workflow JSON with the latest node version, since most mystery breakage is really version drift. Worth remembering too that this whole workflow is built around SDXL's refiner era; a lot of the prompt-weighting fuss here is less necessary now that single fine-tuned checkpoints handle style well on their own.