Subject Styler (Advanced)

The Advanced take on SubjectStyler: the same broad subject categories - Animal, Art, Building, Child, Female, Food, Male, Nature, Technology, Vehicle - but shaped for SDXL's dual-encoder plumbing. If your workflow uses a CLIPTextEncodeSDXL node with its separate text_g and text_l inputs, this is the version to drop in.

Quick refresher on why that split matters: SDXL encodes your prompt through two text encoders, OpenCLIP-ViT/bigG and CLIP-ViT/L, and the second one carries a real share of SDXL's control over composition and subject framing. Talking to each encoder directly is the point of the Advanced node.

What you set

• text_positive_g / text_positive_l - your prompt for the G and L encoders. Same text in both is the normal case.
• text_negative - your negative prompt.
• subject - the category dropdown; No option disables styling.
• negative_prompt_to - Both, G only, or L only, deciding where the preset's built-in negative goes. Default Both.
• log_prompt - prints the assembled prompt to the console.

The six outputs are text_positive_g, text_positive_l, text_positive, text_negative_g, text_negative_l, text_negative. Feed _g and _l into CLIPTextEncodeSDXL's matching boxes; if you're on a plain single CLIP Text Encode, take the merged text_positive / text_negative and ignore the split.

Installing it

ComfyUI Manager, search ComfyUI_MileHighStyler. If Manager doesn't list it (it's an older pack and coverage is hit-or-miss), clone it:

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

Restart ComfyUI. No models, no dependencies - just JSON templates, so it's instant to install and free at runtime.

Common snags

Two things bite people. First, the subject categories are deliberately coarse - Male, Vehicle, Nature frame the shot, they don't describe it, so you still need a real subject in your positive text. Second, the six outputs: don't wire all of them, pick a lane. SDXL dual-encode uses _g and _l; anything else uses the merged pair.

A note on picking Advanced over basic: only reach for this version if your workflow genuinely uses a CLIPTextEncodeSDXL node. If you're on a plain single CLIP Text Encode, the extra G/L machinery buys you nothing and the basic SubjectStyler is cleaner. The split earns its keep only when both encoders are actually in play.

The pack-wide reality check: the whole two-encoder design only exists on SDXL, so this node belongs on the SDXL lineage - Illustrious, Pony, Juggernaut, NoobAI. On a 2026 LLM-encoded model (Flux 2, Z-Image, Anima) there's no G/L to split, the style weighting is discarded, and the negative is inert at CFG 1. It's the right tool on the older stack and a no-op on the newer one.