🌟 CR SDXL Prompt Mix Presets

This is Comfyroll's current answer to a real SDXL annoyance: the model wants your prompt split across a G encoder, an L encoder, and - if you're running a refiner pass - a third slot for that. Hand-copying text into three separate boxes every time you tweak a prompt gets old fast. CR SDXL Prompt Mix Presets takes a base prompt and a style prompt, and a preset that tells it exactly how to distribute them across all three slots - and unlike its predecessor (CR SDXL Prompt Mixer, now marked legacy), the presets here are actually named for what they do instead of numbered 1 through 5.

How it works

You write your main subject text into prompt_positive/prompt_negative, and a separate style or refiner-flavored phrase into style_positive/style_negative. The preset dropdown then decides the mix: "default with no style text" ignores the style fields and just puts your base prompt in the G/L slots, nothing in the refiner slot. "default with style text" folds the style text in too. "style boost 1" and "style boost 2" push the style text in more aggressively - two different strengths of leaning on it. "style text to refiner" routes the style text specifically to the refiner (_r) outputs and keeps the base prompt clean in G/L. The exact internal weighting per preset isn't published anywhere, so the names are your best guide - pick the one that matches the effect you're going for and check the render.

The inputs and outputs that matter

• prompt_positive / prompt_negative - your main subject prompt (defaults prompt_pos/prompt_neg as placeholders).
• style_positive / style_negative - the secondary style text (defaults style_pos/style_neg).
• preset - the five named mixing modes described above. This is the field that actually decides your output, so it's worth trying more than one on the same prompt pair.

Outputs: pos_g, pos_l, pos_r, neg_g, neg_l, neg_r (all STRING) - wire the G/L pairs into your base-pass encoder (like CR SDXL Base Prompt Encoder) and the R pair into your refiner's text encoder if you're running one - plus the standard show_help link.

How to install it

Comfyroll Studio (Suzie1 and RockOfFire), no models or heavy dependencies.

• ComfyUI Manager - search "Comfyroll Studio," install, restart.
• Manual - cd ComfyUI/custom_nodes && git clone https://github.com/Suzie1/ComfyUI_Comfyroll_CustomNodes.git, then restart.

Common issues & troubleshooting

No refiner pass in your workflow? Perfectly normal - leave pos_r/neg_r unconnected. The SDXL realism community largely moved past the refiner stage once fine-tuned checkpoints matured; treat it as optional, not required plumbing.

Style text not showing up in the output at all. Check your preset - "default with no style text" deliberately drops the style fields entirely. If you want your style prompt to actually influence the result, pick any of the other four.

Not sure which preset to reach for. Start with "default with style text" for a normal single-pass workflow with no refiner, or "style text to refiner" if you specifically want your base subject kept clean and your styling isolated to a separate refiner pass. The "boost" presets are for when the default mix reads too weak and you want the style text pushed harder without rewriting it.

Comfyroll nodes missing entirely. Check your startup log for Comfyroll Studio: Failed to load Graphics nodes followed by NameError: name 'CR_HalftoneGrid' is not defined - a symptom of an import failure earlier in the pack, commonly a Pillow version conflict from another custom node taking the whole load down. Force-reinstall Pillow (pip install --upgrade --force-reinstall pillow) or reinstall the pack cleanly and restart.