AC_Third_Season提示词

The third season of ComfyUI-AC_FUNV2.0's prompt library has a type, and the type is cyber-girl. AC_Third_Season提示词 ships 100 prompts dominated by 赛博 (cyber) mecha girls - 赛博机甲少女, 超梦幻赛博机甲少女, MIX赛博少女, 赛博女枪神 - plus a healthy serving of angel themes (天使之翼少女, 炽天使, 堕天使) and a few outliers like 国风长安少年 (a Chinese-style Chang'an youth) and 红发旗袍女孩 (red-haired qipao girl). If your workflow is armored anime girls, this is your season.

The mechanism is the pack's one constant: ModeList is a dropdown that lists every .txt file in the Thirdly folder (yes, the folder is called "Thirdly" - the author's folder names don't always match the node names, which is worth knowing when you go hunting for the files on disk). Select a file, the node reads it, and the contents arrive on the single STRING output. Wire it to a CLIP Text Encode and generate.

Same honest caveat as the other seasons, worth repeating because it's the difference between a good and a bad first run: these files are SDXL-era tag syntax - ((masterpiece)), best quality openers, (cyberpunk1.3)-style emphasis, Danbooru-flavored vocabulary. They're tuned for the anime-capable SDXL finetunes (Illustrious, NoobAI, Pony) that understand 1girl and booru tags natively. Point them at a natural-language model and you'll get something, but you'll be leaving most of the prompt's intent on the table. The mecha/cyber subject matter in particular tends to render best on models that were trained on that vocabulary.

The input that matters: ModeList. One dropdown, required, nothing else. Pick a file name, hit run.

Install is the pack standard:

cd ComfyUI/custom_nodes
git clone https://github.com/A719689614/ComfyUI-AC_FUNV2.0

restart, then under ♑AC_FUNV2.0. No models to download, no dependencies beyond stock ComfyUI - the pack's requirements.txt pins torch==2.11.0 but that line is noise; skip it if pip complains.

Practical notes. First, these are character prompts, not scene prompts - most start from a subject and build the look around it, so you'll usually want to add your own background or environment clause to the end rather than run the file as-is. Second, the file names are the only map you get: 100 Chinese descriptions in one dropdown, and the node won't show you the contents until it runs. If you want to preview, wire the output into the pack's Panel node and read it on the canvas before committing a generation. Third, you can grow the library yourself - drop .txt files into the Thirdly folder and restart, and they join the dropdown.