Full Game Beta Release!

Hi everybody,

I've finally gotten around to finishing this game--at least a first pass at it.

Ever since seeing the La Blue Girl OVAs many years ago I've thought sex combat was an interesting and underutilized trope. This game is an attempt to translate that into the engine I'm building. This has been an opportunity to incorporate a bunch of new functionality, including:

• Complex NPC logic to complement the LLM generated narration. Every fight is highly curated while still allowing for open-ended interaction, unlike the interactions in my previous games, which have all been almost completely emergent and player driven.
• RPG combat style interactions with stats and resources being managed by game logic rather than the LLM.
• Player-defined abilities whose power is constrained by the narrator, allowing for player freedom while maintaining game balance

My hope is that the game allows players to have interesting and creative interactions with the NPCs, co-creating scenes they never would have thought of just messing with a chatbot.

By the way, I've also released a huge update to Orgone Collector.

Relative to the demo, the full version includes:

• 8 opponents vs 3 in the demo. 
• Total overhaul of the LLM backend: I've updated the game from Command-R+ to the Command-A model, and the LLM is now prompted with a more complex reasoning process, improving the narration and making the narrator more firm in reining in the player when necessary. You should also see less repetitive narration than before. Further,  the switch to Command-A finally allows for a longer chat history to be maintained (optional--it slows the API calls down)
• Total overhall of the image engine: a more sophisticated image search powered by a custom, NSFW friendly local embedding model and using a second reranking step using image captions rather than tags to capture more complex semantic information. Also, proper GPU acceleration has been added for the local model, making it about 10-100x faster with appropriate hardware.  Additionally, Gelbooru changed their image hosting, so I've revised the image handling appropriately. Images should load more consistently now.
• Dynamic reaction images. Unlike Sketchy Mind Control App, where these images were AI generated and selected by the LLM, I've greatly streamlined this feature to make it scaleable: images are sourced from Gelbooru, as with the narration images, preprocessed to find faces, and the image is loaded and dynamically cropped to the character's face at runtime. They're also selected using the embedding model--comparing the narration to predetermined descriptions of each expression. This takes the load off the LLM and still yields excellent results. This approach does mean that there are fewer expressions than in SMCA, however, as not all expressions have common gelbooru tags identify them.
• Lots of QoL improvments: autosave on victories, animated healthbars, more verbose output from the game engine to clarify what is happening and  to help with LLM steerability.