Anime Character Prompt Selector (V2)

Anime Character Prompt Selector (V2) is the main node in the Azazeal Anime Characters pack, and the name is only half a lie - it does select characters, it just doesn't call any API to do it and needs no key. Pick an anime from a dropdown, name the character, and it hands you a ready-to-use prompt string for that character. No model downloads, no network calls, no training. It's a local lookup over a bundled catalog of ~15,800 anime.

Why you'd reach for it: character consistency is one of the oldest unsolved problems in local generation - the model has no memory of the character you generated last run, so your only lever is a prompt that names the character precisely. This node gives you that name (and the series context) without you having to remember or mistype it. You can spend your mental budget on the image, not on getting "Makise Kurisu" spelled right.

How it works

The pack ships ~16,000 JSON files, one per anime, each mapping character names to prompt strings. On startup it builds a lightweight .index.json (anime title → file path → character list), and only parses a full prompt file when you actually run a title - loaded lazily and cached in an LRU so switching between shows doesn't re-read everything. It's a genuinely nice bit of engineering: the older 1.0 design registered one ComfyUI node per anime, which made startup crawl and buried the node menu under thousands of entries. V2 registers exactly two nodes and loads data on demand.

One honest caveat: the "prompt" is exactly Name from Series, e.g. Ayukawa, Hana from NegaPosi Angler. That's natural language, not a Danbooru tag list. On modern anime checkpoints that understand the from <series> convention it works fine, but don't expect it to be a full character-LoRA. Treat it as the identity anchor of your prompt, then extend it with quality tags, pose, and style yourself.

The inputs that matter

anime_title - dropdown with ~15,804 choices. It's a big list; start typing and ComfyUI's combo box filters it.
character_name - free-text STRING (default ""). Type it exactly as it appears in the catalog - matching is exact and case-sensitive. If you'd rather not guess, wire this input from the pack's AnimeCharacterListV2 node.

The outputs

character_prompt (STRING) - the one you want. Wire it into the positive input of a CLIP Text Encode node (or a prompt-concat/text node if you want to append style tags).
character_name (STRING) and anime_title (STRING) - the values are echoed back out, which is handy if you're building prompts programmatically or logging what you generated.

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/azazeal04/Azazeal_Anime_Characters_ComfyUI

Restart ComfyUI. Or via ComfyUI Manager → Install Custom Nodes, search anime_character_selector. Zero Python dependencies (no requirements.txt, stdlib only), no model files, no API key - the only weight is the ~88 MB of JSON data that clones with the repo.

Common issues

Empty character_prompt output. That means the character_name string didn't match any entry in that anime's file - usually a case or spacing mismatch ("Kurisu" vs "Makise Kurisu"). Paste the exact name, or use the List V2 dropdown output as the input.
Old workflows show missing AnimePromptNode_* nodes. The legacy per-anime nodes are disabled by default for startup performance. If you load a 1.0 workflow, either set the env var before starting ComfyUI:
```
AZAZEAL_ENABLE_LEGACY_NODES=1
```
or run the pack's bundled scripts/migrate_workflow_v1_to_v2.py to convert the workflow to V2 nodes.
First lookup for an anime feels slow. That's the lazy loader kicking in - subsequent runs are cached. Not a bug.

It won't fix character identity by itself, but for a specific named character on a model that knows the name, it's the difference between a garbled guess and a name the checkpoint actually recognizes.