Improved TTS
All users across all tiers can now enjoy vastly improved TTS logic:
- LLM-generated script: Uses an LLM to generate which voice says what, which is vastly superior to naive algorithmic matching based on punctuation.
- TTS options: Choose from single-voice narration, narration with dialogue (abridged, omitting text like “he said”), narration with dialogue (full), or dialogue only.
- Voice matching system: Each NPC now gets a “voice profile” which can be matched to a specific voice in your chosen TTS service as defined in the .json file under the TTS folder. These include gender, approximate age, and tags such as “deep” etc.
Sapphire users now have a new TTS option that many might prefer over Tiktok TTS.
Added a custom TTS option for you to specify your own service URL. Voice IDs may be specified for a quick setup, but for a full proper setup, you should configure your .json file for voice matching as described above.
Added a global speed multiplier available for all TTS options except Tiktok.
Obsidian subscription tier
This new tier features double the context size of Sapphire, a significantly better image generation model, much higher limits for grid/panoramic mode, and superior TTS in multiple languages.
Removed Wombo image gen
Wombo recently overhauled their website and added a daily user limit of 5 per day, which broke Wombo generations. Although it still seems to allow infinite generations (5 per session with unlimited sessions), it would feel like an exploit that goes against the intended usage of their service, so I’ve removed official support for Wombo.
Regenerate high-quality image (10 per day)
To alleviate the loss of Wombo image gen, all users now get an extra option below “Regenerate image” called “Regenerate image (HQ)”. It will use the default model from the tier above your current tier (for Obsidian, there’s a special model only used for HQ generations). This can be used up to 10 times per day. Please note that after a certain point image quality may be subjective, so HQ doesn’t always guarantee a better image.
Usage-based skill XP in Immersive mode
Due to popular demand, added an option for usage-based skill XP gain in immersive mode. The skill caps are the same as before, but the skills will gain XP by using them during roll checks rather than by investing skill points. You may also switch between these two systems, but be aware switching skill modes mid-run is not well-supported and may affect game balance.
Misc
- Focus input field hotkey: Added a hotkey to focus the input field (defaulting to “/”).
- Better faction images: The LLM will now generate a prompt specifically for the faction icon. You can also specify in Options whether to use the faction image as the icon when viewing the NPC.
- Improved URL detection logic for custom endpoints, including support for “/v2” or other version numbers other than v1.
- Added some more mitigations for parsing malformed AI JSON output.
- Added world background for recipe-based crafting to prevent out-of-universe items being suggested.
- Enhanced moddability of full prompt string.
- Enhanced user customization for scenario context limits (and adjusted to fit token limits). These can be adjusted in text generation options.
- Fixed ruleset updates sometimes getting cut off resulting in an error.
- Simplified lorebook algorithm; it is now based simply on the latest matched entities in text and player input.
- Added region opacity options.
- Added option in Stripe subscription page to upgrade or downgrade your subscription. In this case, your fee will be prorated based on the remaining time of your previous subscription.
- Fixed issue where AI would sometimes ignore player instructions when spawning entities due to the anti-anachronism instruction.
- Fixed bug where corpses can die again from delta hp.
- Reduced chance of Unified AI gaslighting user about game state, and added instruction about player request taking precedence over other instructions in Unified mode.
- Maybe fixed infinite quest completion issue with V4 (story arc) quests.
- Increased Sapphire/Obsidian allowance for story turn completion length.
- Tweaked story length for travel turns to reduce chance of story getting cut off.
- Fixed bug where player input wouldn’t trigger lore keys in Unified mode.
- Improved lore key triggering for items; item will also be included in inventory context when triggered.
- Fixed bug where user could edit an untracked quest and produce an error.
- Added button linking to config folder for adjusting timeouts for custom text generation.
- Overhauled GENERAL_QUESTION_ANSWERER which was an overused preamble which polluted a lot of prompts unrelated to generations, and migrated the parts specific to entity generation to ENTITY_GENERATOR.
- Tweaked place generation prompt to reduce chance of duplicating pre-existing entities.
- Added default specialized narrative rule to prevent repetitive generations.
- Fixed issue where crafting plausibility wouldn’t respect scenario settings plausibility mode.
- Added base attribute and base skill number (before item/faction bonuses) for tooltip when hovering over attribute/skill.
- Made corpse duration configurable in Options. Also increased default num turns as corpse and prevented counter from decrementing for system turns.
- Fixed issue where Unified could only do short side quests. It now presents the dropdown with all 4 quest types to pick from.
- Added new side quest button, in quest view.