Bug fixes:
- characters.json had a trailing comma, so it wasn't valid JSON: Character
Select listed no characters, and Character Vault treated the file as empty
and overwrote it (losing every saved character) on the first save. Fixed
the file, and Character Vault now refuses to save over an unreadable
database instead of wiping it.
- Scene Director loaded Scene_Director.md, which isn't shipped; it now uses
the bundled Scene_Director.txt unless a .md is present.
- web/index.js imported ./js/creepynodes_appearance.js, which doesn't exist
in this repo. The extension is also renamed so it no longer clashes with
Creepy_nodes' "Creepynodes.appearance" when both packs are installed.
The console test script moved out of web/ so it isn't auto-loaded.
- Gemini API key config pointed at a placeholder C:\ path; it now defaults
to the bundled assets/scripts/api_keys_config.json with relative key
paths, and GEMINI_API_KEY is supported.
Gemini (same fixes as Creepy_nodes 2.10.0): thinking "disable" actually
disables thinking on 2.5 Flash (it was eating max_output_tokens), the
seed is sent to the API, empty/blocked responses report why, current model
list plus a custom_model field.
Also: prompt files are read at run time and re-run when edited, character
list refreshes without a restart, LoRA dropdown uses ComfyUI's cached file
list, __init__ isolates per-module import failures, and requirements drop
unused pins (importlib_metadata, packaging, opencv-python, sounddevice,
standard-imghdr).
Version 1.3.0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_013Lm4YDXhwj49hAA1RKS58Q
Updated prompt generation guidelines to include image input analysis and detailed prompt structure for both text and image inputs. Emphasized clarity, detail, and style flexibility while allowing for NSFW content in prompts.
Claude