APNext Character

This is the fan-art engine of dagthomas's "SDXL Auto Prompter" pack. It's a wall of dropdowns full of named characters - 374 fictional characters, 168 anime characters, 100 superheroes, 97 supervillains, plus monsters, fantasy archetypes, sci-fi roles, action types and a "general" list of human archetypes - and it appends whatever you pick to your prompt. If you spend your time doing character illustration, cosplay-style renders, or "what if X met Y" mashups, this saves you from remembering exact spellings and gives you a randomizer for free.

Here's the thing worth knowing: this node punches above the rest of the family specifically on anime and illustration models. Illustrious, Pony and NoobAI are SDXL finetunes trained on Danbooru-tagged data, and Danbooru tags include character names. So "asuka langley soryu" or "aragorn" isn't a vague style hint on those checkpoints - it's a tag the model was actually trained on, and it'll often nail the character's design. On a general photoreal base, the same name is much weaker.

How it works

No model, no API, no key. It concatenates your prompt with the character(s) you select, joined by the separator (comma default), and returns a string. Each menu leads with None / Random / Multiple Random, so a seed lets you roll random characters - genuinely fun for batch mashups. It's tag-style injection, which is why it lands so well on the SDXL/booru lineage and gets fuzzier on LLM-encoded 2026 models.

The inputs and outputs that matter

• prompt - your base text; character names get folded in.
• The menus you'll actually use most: anime (168), fictional (374 - the broad movies/games/books cast), superheroes (100) and supervillains (97). monsters, fantasy, scifi, general and action cover archetypes when you want a kind of character rather than a specific one.
• seed - drives the Random / Multiple Random picks.
• attributes (off by default) - adds the extra descriptor words the pack attaches to entries.

Outputs are prompt and random, both strings, wired into a CLIP Text Encode.

How to install it

ComfyUI Manager: search comfyui_dagthomas, install, restart. Or:

cd ComfyUI/custom_nodes && git clone https://github.com/dagthomas/comfyui_dagthomas
cd comfyui_dagthomas && pip install -r requirements.txt

then restart. The requirements.txt pulls the whole pack's dependency stack - transformers, decord, scipy, openai, anthropic, google-generativeai - because the repo also ships cloud-LLM and video nodes. This builder needs none of it; decord is the usual culprit if the install errors, and the plain nodes generally load regardless.

Where people get burned

• Wrong model for named characters. On a photoreal SDXL or an LLM-encoded base, "all might" may produce a generic figure. Run character prompts on Illustrious/Pony/NoobAI where the names were trained in.
• Copyrighted designs. These are trademarked characters. That's fine for personal exploration; think before you publish or sell.
• Mashup mush. Two characters at once can blend into a chimera. If you want a clean crossover, generate them in separate regions or passes rather than dumping both into one prompt.
• Randomize responsibly. Multiple Random across the 374-deep fictional list is a great idea machine and a terrible way to get a specific result. Use it to explore, then lock the pick.