I'm very excited about this update because it's something I have found lacking in other AI games which try to slap on an attribute system that doesn't actually tie into the game mechanics.
In this game, your character attributes will actually influence the outcome of your actions by biasing the AI. It works by having the AI figure out which attribute the roll check should be on, and then the game feeds a positive or negatively biased action string to the AI based on the result of that roll.
Possible results:
Critical failure: "You try to _, but"
Minor failure: "You try to _"
Minor success: "You _"
Critical success: "You expertly _"
So far, attributes only influence the outcomes of quest completion actions (completed via the "action choices" option). This goes hand-in-hand with the quests v3 update and was actually in the original vision of Quests v3.
The attribute detection still needs some work as some attributes are more OP than others, but this will be tuned over the following weeks.
Compatibility with Steam Deck / Proton
I have implemented a workaround to the "SetFilePointer" Proton bug. You can now play the game with Steam Deck / Proton with minimal settings. The issue of not being able to use AMD as the GPU still exists; however, using the smallest AI model in tandem with cloud text generation will still deliver good results.
Quests V3
Quests are now slightly more epic than before, with each quest consisting of multiple stages.
When you first begin your quest, there will be multiple approaches you can choose from.
Once you've chosen your 3-stage approach, you can complete each stage via the same mechanic as before (item/ability), or via a new mechanic called action choices. The action choices presents you with multiple ways to complete the task, each of which will influence the outcome of the story in its own way.
Location state changes
Location state changes: Entities can now have state changes (different from status effects). In all likelihood, status effects will be sunset in favor of state changes, in the future. So far, only a location can have a state change. For example, if you cast “flames” inside a building, the AI could determine that the building is now on fire. Then, the game will automatically regenerate the image to reflect the current state of the location.
Misc
- Implemented parallelized calls for cloud generation. That means for cloud and novelAI generation you might see some lower generation times especially when visiting a new location. Please note however that currently the response times are very finnicky and could be slow regardless (this is a temporary issue) - Fixed issue with disallowed words with krake models.
Slightly improved cloud generation
Improved cloud generation speed significantly, and made it slightly less wacky. There are still some known issues with accuracy and weird names, which will be ironed out after a while.
Also fixed a bug in map display where it wouldn't get the correct root location if you were in a non-starting zone of the root location.
World Map Update
There is finally a world map to view all your locations simultaneously and the option to fast travel to any of them. This makes traversing and orienting oneself with the world much easier.
The location hierarchy has been slightly simplified such that it is no longer possible for each "zone" of a location to have a different parent location.
Misc: - Probably fixed bug where image gen failed for users with computers in a foreign language - Fixed sound volume bug when exiting combat - Changed default sort order to chronological - Flashing red warning when area level is too high relative to player level
Optimized event checking
I finally delve into an old question I posted in github and implemented the "right" way to do probability checking. It's about 5-10x faster than before.
Your GPU will no longer sound like "tick tick tick" when checking probabilities; it'll be more like "brrrr", like the story text generation but much shorter.
Since story text generation is still the same, overall the improvement in text generation speed may only be about 10-30% depending on your settings. This affects all text generation (both local and cloud) because cloud generation still uses local text generation for probability checking.
Inventory sorting, image generation progress
- Added UI indicator for which images are loading and their progress - Added logic allowing partially generated images to be viewed while the image is generating - Added sorting and filtering options for the inventory - Changed exploration view to a tabbed view so all the images/entities can be clearer - Added ability to drop/abandon item - Fixed some tooltip display issues and added some tooltips - Added a partial fix for the Wombo NSFW issue (gives up generating the image if Wombo doesn't return anything for a while, so it won't keep asking infinitely)
Configurable event triggers
Event triggers are now configurable via some sliders in the text generation options menu. All the way to the right means always on (as long as it's a valid event for that action); all the way to the left means it'll never happen; middle is default as determined by calibration.
Charm update
- Added the "charm" check to the AI event checks. An enemy can now be charmed which qualifies as instant defeat (except that they are still alive, and become an NPC)
- Updated Wombo settings to include their new settings which use Stable Diffusion; changed default to “realistic”, which looks wayyy better than the previous setting, and almost as good as locally run stable diffusion.
- Fixed stable diffusion bug where it crashed for certain characters (hopefully actually fixed this time)
