Essences: Essences allows you to perform manipulations in the game such as changing an entity, editing the last story turn, and taking an “open-ended” turn.
Essence of Transforming: Allows editing name/description of 1 entity
Essence of Rewinding: Allows editing the last story turn
Essence of Opportunity: Allows 1 “open-ended” turn
Essences have a chance of dropping from regular mobs, at least 1 guaranteed drop from an elite mob, and 1 of each will drop for a legendary boss. Your max # of essences is influenced by your character level.
Creative mode:
In creative mode you can do all these things infinitely without needing to consume essences. This can be toggled on or off at will in the options.
Misc:
There was an issue where the save would be corrupted if your follower became an enemy and died in a place they weren’t born. This is probably fixed now.
125M viability (sort of)
With this patch, those who don't have a GPU with enough VRAM and don't want to get a cloud generation subscription can have a slightly "less bad" gameplay experience than before. The 125M model has been tested more extensively with event checks and prompt engineering to work as intended at least some of the time.
Full list of changes:
- Autodetection of GPU and VRAM, only on the first run of the game since this update. If not enough VRAM is detected, the text generation will be set as 125M and image generation set to Wombo. - Improved 125M model’s detection of equipment type and consumables - Improved 125M model’s generation of entity names and item names. Item names are now at least sometimes reasonable. - Reinstated “location change” event. The player is now prompted whether to go there, so you can dismiss it if you want. The location will spawn as a sibling location, so you have to go “up one level” to see your original place. - Fixed stable diffusion bug where it would crash for certain characters - Edit details and redo image now available for abilities too
9/8/22
Made the bad words filter option less confusingly worded. Ticking the checkbox means you’re filtering out the bad words. Leaving it empty means you’re not filtering anything.
Speculative: Hopefully fixed bug in stable diffusion regarding the “can't encode character '\u200b'” bug.
Standalone stable diffusion: You can use stable diffusion outside of the game. Updated the txt2img.exe bundled with the game under the “ai” folder and tested that it works. Example command: txt2img.exe --prompt "(Ultra HD Airbrushed Oil Painting) Skaar warrior: Skaar warrior in full leather armor wielding a sword with a triangular blade. The warrior puts his back against a wall." --plms --W 512 --H 512 --n_iter 1 --n_samples 1 --ddim_steps 80 --seed 47
Edit details: You can now edit the name and description of game entities.
Regenerate image: Regenerate image and edit details are now assigned to right click. Also expanded this feature to include enemy and item images.
Possibly fixed discrepancy with line breaks. There was an inconsistency in line breaks between cloud vs local text gen. Now line breaks should be allowed for most actions but disabled during combat (in theory, this gives more variety in general but also prevents combat from becoming too sidelined).
Disallowed words filter, Regenerate bad images
- Added customizable NSFW filter - Fixed bug where “prefix” for VQGAN or Stable Diffusion wouldn’t be applied - Fixed a GPT calibration bug which prevented 6B model from working - Hopefully improved Krake generation for NovelAI users (untested) - Added ability to regenerate images for player character, NPC’s, and current location image.
v1.11: Stable Diffusion
Image generation has been taken to the next level for those lucky enough to own a GPU with at least 8GB VRAM. Unfortunately due to the high VRAM requirements, Stable Diffusion will barely fit on an 8GB VRAM GPU via the "Budget VRAM" option. The good news is it's fast, and if you have enough VRAM to run 512x512 it will look absolutely stunning.
256x256 looks like crap no matter how many iterations you do, and 384x384 is a middle ground.
I've looked into cloud options for Stable Diffusion which are a no-go right now at around 1 cent per low-quality generation which would add up quickly.
Misc: - Added an option to customize the prefix for local image generation - Removed the 2.7B text generation model from the pre-bundled default models to save disk space. It can still be downloaded in-game.
Better cloud/novel AI
Cloud AI fix: I figured out why the cloud-based generation and NovelAI generation were so ridiculously verbose. As it turns out they’re more sensitive to repetition_penalty which was too high for them. Lowering it for the cloud AI’s improved the quality a lot.
Misc: -Fixed issue where load game wouldn’t work if the game was downloaded a long time ago -Hopefully fixed issue where follower damage was incorrectly calculated as unleveled
NovelAI integration, Wombo style options
NovelAI integration: In addition to the regular cloud generation or local generation modes, there's now also the option to use your own pre-existing NovelAI subscription for cloud text generation. Please note, the NovelAI and $1.99 / month cloud generation options are currently not behaving as well as local text generation, due to prompt engineering issues. This will be improved over time.
Wombo.art style choices: Added option to pick style in Wombo.art image generation
Added ability to abandon follower as a dropdown menu option. It will drop them off at the current location.
Slightly increased text size of debugging screen
cloud-based text generation
There's now an option to use cloud-based text generation if you don't have the right kind of GPU to run the AI locally. Currently the price is $1.99 per month. To use it, select "cloud-based" under text generation options, subscribe via the web page, and copy/paste your customer key into the text field.
One known issue is that the generated entities are too verbose. I think I have a decent idea on how to solve it. I need to also pass the disallowed words into the cloud-based API to prevent the names from being too long.
7/29/2022
Added ability to save/delete custom world background presets
By popular request, “[OPEN-ENDED]” conversation option is now always available instead of only being available in Adventure difficulty
Fixed bug where in-game music volume wouldn’t remember the player’s preference correctly
Fixed issue where game would be unresponsive with Wombo generation when closing or pressing Alt+F4
Fixed issue where Wombo generation would sometimes fail
hotfix
Fixed a bug from the previous patch where consulting AI could sometimes be stuck forever Fixed a bug from previous patch where Wombo image gen might fail Fixed rate-limit logic for Wombo
