Mood Styler (Advanced)

Same job as MoodStyler - dial in the emotional tone of an image from a dropdown of Aggressive, Calm, Chaotic, Energetic, Happy, Mysterious, Relaxed, Romantic, Sad, Serene - but built for SDXL's two-encoder setup. If your workflow feeds a CLIPTextEncodeSDXL node with its separate text_g and text_l inputs, this is the version that slots in without an adapter.

The reason there's a split at all: SDXL encodes your prompt twice, through OpenCLIP-ViT/bigG and CLIP-ViT/L. That second encoder carries a real chunk of SDXL's control over tone and composition, so the Advanced node lets you talk to each one on its own terms instead of duplicating a string into both.

What you set

• text_positive_g / text_positive_l - your prompt for each encoder. Same text in both is fine and normal.
• text_negative - your negative prompt.
• mood - the preset. No option disables it.
• negative_prompt_to - Both, G only, or L only, controlling where the mood's built-in negative is routed. Both is the default and usually what you want.
• log_prompt - prints the assembled prompt to the console so you can see what went into G versus L.

Outputs cover everything: text_positive_g, text_positive_l, a merged text_positive, and the three matching negatives (text_negative_g, text_negative_l, text_negative). Send _g and _l into CLIPTextEncodeSDXL. Not on an SDXL dual-encode graph? Ignore the split and grab the merged text_positive / text_negative for a plain CLIP Text Encode.

Installing it

ComfyUI Manager, search ComfyUI_MileHighStyler. If it doesn't list (it's an older, niche pack and Manager is spotty about it), clone it:

cd ComfyUI/custom_nodes
git clone https://github.com/TripleHeadedMonkey/ComfyUI_MileHighStyler

Restart ComfyUI. No models, no dependencies, no requirements.txt - just JSON templates, so it installs in seconds and uses no VRAM.

Common snags

Mood is a soft effect, and on a busy, heavily-specified prompt it can get drowned out - it lands best on portraits and clean compositions. If you're not sure it did anything, log_prompt shows you the truth. And the six outputs trip people up: pick a lane. SDXL dual-encode uses _g and _l; anything else uses the merged pair.

When to pick this over the basic MoodStyler: only if your graph actually feeds a CLIPTextEncodeSDXL node. On a plain single CLIP Text Encode the G/L split is dead weight and the basic node is tidier.

One honest limit worth repeating: this is CLIP-era tooling, and the whole two-encoder premise only exists on SDXL. On an LLM-encoded 2026 model (Flux 2, Z-Image, Anima) there's no G/L to split, the weighting is discarded, and the negative is inert at CFG 1. So MoodStylerAdvanced is a great fit on Illustrious, Pony, Juggernaut and friends, and a non-fit on the newer stuff.