Fantasy Styler

Fantasy prompts get long fast - "a wizard in a tower surrounded by ancient tomes, glowing runes, candlelight" - and then you still have to type it again for the next scene. FantasyStyler puts 52 fantasy scene presets on one dropdown: Alchemy Lab, Dragon's Lair, Eldritch Horror, Enchanted Forest, Magic Carpet Ride, Sorcerer's Tower, Wizard's Library. Type a subject, pick a scene, and the node builds the rest of the prompt plus a matching negative.

What it actually is

One of the per-theme stylers in aegis72/comfyui-styles-all, the pack that preserved azazeal04's removed SDXL Prompt Styler styles. The mechanism is dead simple: each Fantasy option is a JSON template with a {prompt} placeholder and a negative prompt. The node substitutes your subject into the placeholder and appends the style's negative to yours. Text in, two strings out, nothing generative in between.

The menu reads like a campaign setting catalog - Celestial Realm, Dwarven Forge, Elven Kingdom, Mermaid's Song, Moonlit Ritual, Mythical Beasts, Sword of Destiny - which makes it genuinely handy for keeping a consistent world feel across a character sheet or a storyboard of scenes.

The inputs that matter

• text_positive - your subject; the scene wraps around it.
• Fantasy - the 52-option dropdown; None passes through.
• log_prompt - on by default; prints before/after to the console.

Outputs: text_positive (styled) and text_negative (your negative plus the style's, comma-joined). Both STRING - straight into CLIPTextEncode.

Installing it

ComfyUI Manager (search ComfyUI-styles-all), or:

cd ComfyUI/custom_nodes
git clone https://github.com/aegis72/comfyui-styles-all

Then restart ComfyUI. No dependencies, no models.

Where people get burned

• Negative stacking. Every preset adds its own negative; keep your own negative box short or the combined string gets bloated.
• Restart after edits. The dropdown loads from data/Fantasy/Fantasy.json at startup.
• SDXL-era keywords. Great on SDXL-lineage checkpoints (including anime merges), noticeably weaker on Flux and LLM-encoded models.