devlogs.

image generation:

→ prompts now adapt to the model: CLIP models get tags, while Qwen/Flux/Chroma/Anima and others get full sentences — the scene director (pose, camera, lighting) writes prose for them

→ the prompt limit for these models has been raised from 900 to 1800 characters — scenes no longer get cut off mid-word

→ the “Mielgo-style” look is no longer mixed into local models

→ ComfyUI generation timeout increased from 2 to 10 minutes — heavy models now have time to finish

→ three new ComfyUI workflows: Qwen 2.1, the same with a face detailer, and a faster detailer variant. the multi-reference face generation workflow has been tweaked

→ first-person mode. if you turn off “describe yourself”, you become the camera: only the characters are in the frame, your face never appears. when you take part in the scene, only your hands enter the frame from below (and your body during sex). a kiss is drawn as “the character leaning toward you” instead of two faces pressed together

→ different camera angles in first-person mode: instead of a “distant wide shot” — a top-down view, through your eyes, with the character in full height

new text-to-speech: supertonic 3:

→ runs on the CPU — no GPU or server needed

→ supports every language in the game except Chinese

→ breathing, laughter, and sighs are voiced

video:

→ fix: transitions no longer interfere with showing generated videos

→ fix: animated backgrounds no longer flicker or freeze because of the cursor effect, and the menu no longer hides under generated videos

→ fix: the “frames” slider now shows the saved value correctly

→ clips are converted to webm instead of heavy gif

→ added ComfyUI workflows for Wan 2.1 text-to-video

→ video timeout up to 30 minutes, clearer ComfyUI errors

→ you can now save generated videos as webm

3d models:

→ fully remade models for Laura, Dina, Jo, Mika, and Tisha

→ modular clothing in 3D: outfits are now put on over the character without reloading the model; new “clothes” button in the menu. characters now have default outfits

→ an extra hairstyle for Laura and glasses for Tisha

→ lipsync fully rewritten

→ reworked talking animations

→ fixed bugs when switching models quickly

→ lighting adapts to the background

→ improved eye tracking of the cursor and a smooth return of the gaze to the player

→ smoother camera zoom in and out during speech

→ updated set of six idle animations, the unsuccessful ones removed

→ fixed idle animations resuming after speech and other animations

→ added a smooth turn toward the player, interrupted by manual camera rotation

narrator:

→ no longer confuses you and the characters with the list of all characters

→ if a character card is assigned to a character, the narrator uses the name from the card

→ narrator lines no longer have a name tag

relationship status:

→ receiving a status no longer interrupts the character’s reply

→ shown on its own line: no name, centered, in green

→ also shown with streaming turned off (it wasn’t before)

→ the name from the character card is now used instead of the default character name

character cards:

→ fix: if all base contexts are turned off for a card, the character still gets the “everyone is over 18” rule

→ fixed the “don’t limit AI response length” setting (it wasn’t applied if the character had triggers disabled)

→ fix: cards now fully work in Tisha’s slot with any set of contexts. previously, with checkboxes turned off, she lost her history and part of the context

reworked the “character starts the dialogue” option:

→ opening scenes are no longer the same; the character remembers where you left off

→ if you’ve already talked — they’ll pick up the conversation and take into account how much time has passed

→ works even with history saving turned off — the character remembers the last few lines

→ new acquaintances are always different, recent variants don’t repeat

→ character cards now start the dialogue with their own greetings, if the card has any

✦ support for gpt-6 and newer. if the provider says a parameter isn’t supported (temperature, top_p, etc.), the game drops it and retries the request on its own

✦ gallery: 14 new images: Mika +4, Tisha +3, Kas +3, misc +4

✦ the help screen has been redesigned: cards, icons, quick start; new FAQ section with 12 common questions about API, local models, image generation, and saves; updated list of recommended models

✦ tooltips rewritten

✦ clicking the new version on the main screen now opens the devlog page on the website

✦ updated whisper (speech recognition)

✦ koboldcpp updated to 1.121

✦ saiga nemo is back as the default text model

✦ the crypto payment button is hidden (nowpayments stopped accepting payments because of my passport. nobody used it anyway)

sorry in advance for spamming you with updates lately — the misaligned elements in the settings were really bugging me, and I couldn’t stop tinkering with the new image generation •ᴗ•

🛠 image generation:

→ you no longer need to specify your age or gender in “describe yourself” — they’re taken from your date of birth and the main screen settings. if the field already contains your age/gender, duplicates are automatically removed from the final prompt

→ new “don’t use” toggle next to the “describe yourself” field: excludes your appearance from generated images and fades out the field. useful if your features start bleeding into the character

→ when generating multiple images at once: the first shot is framed however the “director” intended, the second is a close-up, and the third is a wide shot. previously, images in a series mostly differed only in noise/grain and ended up looking very similar; now they’re different shots of the same scene

→ the tooltip for the “describe yourself” field has been rewritten and now explains all of the above

→ comfyui: added support for PrimitiveStringMultiline / PrimitiveInt nodes for Krea Flux models

🛠 fixed the layout of the API models section on the settings page

🛠 added a progress bar when unlocking/locking all gallery items

✨ new gallery items for Tisha, Finn, Kas, Fiona, Laura, and Dina

🛠 local text models:

→ fixed reasoning, system instructions, and service tokens like user / assistant appearing in character replies

→ improved compatibility with Qwen, Mistral, Gemma, and other GGUF model families

→ both regular and streaming responses are now more stable

→ fixed connection to remote KoboldCpp instances

→ bypass eos is now disabled by default; cases where it is actually needed have been reduced to a minimum

→ KoboldAI has been deprecated — it hasn’t received updates for two years, and its own developers recommend using KoboldCpp instead

🛠 image generation models:

→ your self-description now reaches the model completely untouched

→ more room for descriptions: “describe yourself” — 180 instead of 90, character appearance in character cards — 450 instead of 270

→ prompts are no longer blindly truncated. the limit now depends on the model, and if space still runs out, styling gets trimmed first instead of the actual scene: pose, action, and camera angle are never discarded

→ scene generation is now more accurate. the model separately determines who is in the frame, what is happening, the shot type, and the lighting — instead of processing it as one continuous line

→ fixed counting people in the frame: extra people should no longer appear, and required characters should no longer disappear

→ character appearances are preserved better between characters, with less feature mixing and fewer traits from one character leaking onto another. distinctive traits of characters who are not in the frame are now automatically excluded

→ the scene location is no longer replaced by the conversation context

→ fixed cases where, after a failed model response, service text could end up in the image prompt instead of the actual scene description

🛠 changed Tisha’s starting locations in Sandbox

🛠 fixed location synchronization in Sandbox between the text and image models, including the office location

🛠 reworked the way characters are called into a scene from dialogue: this is now handled by a separate bot that knows character card names, understands exactly who was called, and ignores vague requests; duplicated dialogue lines when a character arrives are also gone

! bypass eos is no longer a recommended option. it’s better to disable it, since local models now correctly understand when they should stop generating. with it enabled, the model will keep generating until it reaches the response limit

✨ Tisha’s story event has been completely reworked; u can find it in story navigation under “therapy session (Tisha)”

✨ Tisha is now a full sandbox character (woohoo); you can also visit her for a therapy session in sandbox mode

✨ use jinja - a new toggle in settings. launches koboldcpp with the --jinja flag, so utility requests (image prompts, clothing, detectors, etc.) use the model’s native template

✨ added the ability to generate up to 3 images from a single prompt; the setting has been moved to the “image prompts” section

✨ the game is now fully translated into french, german, spanish, chinese, ukrainian, and portuguese (br)

✨ 4 sweet new transition effects for background changes and generated images

✨ the default text model has been changed to Qwen3.5-9B-The-Defiant-Fable-Uncnr-Heretic-NEO-MAX-MTP-Q4_K_M.gguf

✨ “continue” story button on the main screen

✨ the hyperwoo minigame has been updated to 0.1.35; completing each level now rewards 20–30⦵

✨ Mika (180⦵) and Tisha (135⦵) can now be unlocked in sandbox before they appear in the story

🛠 reworked some of Mika’s weaker story images from the first encounter

🛠 fixed image prompts (and other small utility bots) for certain models, especially gemma

🛠 server-lv, previously used to bypass blocks, has been retired (due to sanctions, beget’s latvian server stopped working); all of its functionality has been moved to server-kz

🛠 changed ua to uk for language AI, in accordance with the ISO 639-1 standard

🛠 claude-sonnet-5 is now selected correctly from the recommended models

🛠 fixed several story dead ends

🛠 redesigned the main screen and context settings. language dropdowns have also been standardized

🛠 koboldcpp updated to 1.119

🛠 added rtx 50-series gpu support for local speech synthesis/recognition and stable diffusion; older gtx 10-series cards (1050–1080) are no longer supported and will fall back to cpu. the same will happen if yr nvidia driver is outdated - version 580 or newer is required

🛠 coqui: torch cu121 → cu130; fixed voice cloning - the reference recording no longer falls back to a preset voice; the voice list also no longer requires restarting the game if u switch to coqui while the game is running

🛠 qwen-tts: torch cu118 → cu130

🛠 updated the coqui and qwen links in settings and help

🛠 whisper and stable diffusion rebuilt with cuda 13.3

🛠 stable diffusion: updated the engine; added support for wan, ltx, qwen-Image, z-Image, chroma, and new samplers; “hot model” now works differently - when enabled, the model stays loaded in vram; when disabled, it stays in system ram, leaving vram free for other tasks; the build is also smaller now; image generation on nvidia gpu has been fixed on linux

🛠 code refactoring

the new text model replacing Saiga Nemo has fewer parameters (9B vs 12B), but in practice it’s smarter, faster, and better tuned for natural roleplay dialogue

generating up to 3 images from a single prompt will be especially useful with local models at a slower game pace, letting each moment play out more fully and making playthroughs feel more varied

gtx 10-series support:

the rtx 50 series requires cuda 13, which no longer supports the Pascal architecture. maintaining two separate builds for different GPU generations would mean several extra gigabytes. according to current Steam statistics, only around 1–2% of users still have these cards, and local models are too slow on them for comfortable gameplay anyway. so, sorry, but I decided to drop them from Stable Diffusion and the local speech models.

as an alternative, koboldcpp supports every GPU generation, so use that if this affects u

linux: this time I built the linux version through github actions, since I temporarily no longer have access to the Ubuntu laptop I used for testing the build. all technical tests pass, but if u run into any problems, let me know and I’ll rebuild it

Tisha: there still aren’t many images for her yet — I’m working on it. there’ll be more, just like with the other characters, and she’ll get her own gallery later. her context also still feels a bit rough to me and there’s definitely room for improvement, so I’ll keep working on it

✨ reworked custom contexts

✨ png/json character cards and loras for custom characters

✨ ava is no longer ava — her new name is mika

✨ story:

→ the mika prison section has been completed and connected to the rest of the story;

→ a new d&d hospital section has been added, with a new “hospital (one thousand and one nights)” point in story navigation;

→ playing hyperwoo with mika: the choice about humanity from level 6 of the mini-game is now saved to mika’s context. she will offer to play it in prison and when you choose her at night, after the park scene with jo, fiona, dina, or finn

✨ a new one thousand and one nights d&d scenario in sandbox mode

✨ tamagotchi: added grok voice support and a voice selector in the tray menu

✨ new music

✨ koboldcpp: contextshift/swa settings are now available directly in the game

✨ android: the game now runs fully edge-to-edge, without those annoying borders

✨ hyperwoo voxel mini-game:

→ the ridge in the intro has been replaced with an arched stone fortress;

→ the ending of level 2 is now a nighttime trail;

→ added comets streaking across the sky on the trail and during the storm intro;

→ lava and the sea of stars now fade smoothly into the fog without a visible rectangular edge;

→ respawning now happens by sector instead of restarting the entire level;

→ improved performance and forced use of the dedicated gpu;

→ android version

🛠 the main game now prioritizes the dedicated gpu, which is especially relevant for laptops and primarily needed for mini-games and visual effects

🛠 changing the lora used by the koboldcpp image model no longer requires restarting the model

🛠 koboldcpp updated to 1.117.1; electron and three.js have also been updated

🛠 android: fixed a story-mode bug during the morning scenes in jo’s, dina’s, and fiona’s apartments that prevented further progress

🛠 collapsed settings tabs now display active models and other enabled features inside them

🛠 improved start screen performance

🛠 character thoughts now disappear after the text field is updated

🛠 rewrote the narrator context

🛠 updated the list of recommended local text models and api models for claude, gemini, chatgpt, and grok

🛠 minor interface improvements and code refactoring

✨ story: added a summary after long free-form stretches; character skill system and a trust meter; reworked the event where Dina offers to play a video game; for ruin 3042 the reward for winning is 40⦵

✨ Uncensored-Local-AI for android

✨ gamepad support

✨ koboldcpp: support for the Anima model; short videos via Wan models; more than one LoRA

✨ you can now type a new line with shift + enter

✨ link to Fiona's instagram in the phone and in the story (prologue, when you sort out your things in the room)

✨ quick access to the minigames added to the main screen — "save the princess", "poker" and "ruins 3042"

✨ new music

✨ sexy images for Dina; images for Ava

✨ items in the gallery

✨ tamagotchi: koboldcpp support, pick the text model via the tray menu

✨ d20 roll modifiers in dnd based on skills

🛠 story: rewrote a small bit with Dina in the hospital; fixed the rooms panel — it now hides when you leave characters' apartments

🛠 koboldcpp: updated to version 1.115.2

🛠 comfyui: you no longer have to re-select the workflow file in the game after changes in comfyui

🛠 ruin 3042: fixed weapon racks; tweaked stun timings on a head jump

🛠 promptchan: proper error output when something goes wrong

🛠 rewrote and shortened the default negative prompt for images

🛠 fixed browser speech recognition

🛠 fixed nvl-mode

🛠 help section brought up to date

🛠 answer options and the d20 roll window no longer cover long text — you can scroll

🛠 for claude-fable-5, claude-opus-4-7/4-8 (and newer) the deprecated sampling and temperature params are correctly ignored, so the small bots don't crash with an error

🛠 fixed a bug with background zoom when switching a 3d character to 2d

🛠 improved detection of when to roll the d20 — the bot no longer makes up a value but waits for the roll

for the skills I drew inspiration from disco elysium (of course), modern analytic philosophers of mind, and also nietzsche, horney (neuroses) and byung-chul han (the burnout society)

the skill and trust system works in both story and sandbox. to reset skills you need to start a new game (after the short plane video). by default every character is at 0% trust. in the sandbox, analysis and counting of skills and trust kicks in after sleeping, on the next day. in the story — after long free-form conversation stretches.

to save on system context, info about skills (at this stage) is given to the bot in a sparse form with a simple list. on the next step, once I optimize the characters' system contexts, skills will affect gameplay more substantially

video generation: works with wan models, short clips up to 5 seconds. in the help section for local image generation, at the end, there are the necessary instructions and links

uncensored local ai (android): the models are very small, the context window too, the full system context doesn't fit, so you'll most likely run into needing to trim the system context and shorten the message history

✨ "grok voice think fast 1.0" speech

✨ story, prologue:

before entering the house and before sleep, added the option to take a walk around the city;

fully rewrote the first day when you fall into the loop;

dinner event with Fiona, Laura and Casper;

poker event;

✨ "multi-dialogues":

a 3rd character on screen;

improved prompts for interaction between several characters;

in a dialogue each one tries to speak only for themselves;

new "stay silent" button — active when there are more than 2 characters on screen;

reworked characters leaving/arriving

✨ new images for Laura, Fiona, Josephine, Dina, Fin and Casper

✨ new items in the gallery

✨ updated images on the start screen

✨ updated 3d models (vrm) of Fiona; added a lingerie outfit; changed eye behaviour when tracking the mouse; increased the model's tilt angle

✨ in the log there's now a button to delete all lines in history below a certain one

✨ music: three new tracks

✨ updated the story editor

✨ updated the help section

✨ poker: added a cheat-sheet with hand combinations to the instructions

✨ redesigned the main screen; color correction and lighting via webgl on the background and characters

✨ added a "prologue, start of the loop" point to story navigation

🛠 redesigned the settings tabs for speech and images

🛠 default image model ilustmix updated to v11

🛠 koboldcpp updated to 1.112.2

🛠 electron updated to 39.8.9

🛠 fixed a small branch if you choose to sleep with Ava on the couch at Fiona's on the fourth day

🛠 opening poker changed to a less expensive circle effect

🛠 headers in story choices are now shown above the options

🛠 fixed a bug with blur on generated images when certain triggers fire

🛠 redrew some poor locations

🛠 fixed the custom cursor arrow

🛠 button to skip long dialogue chains (fixes problem spots)

🛠 android — images reduced and optimized

grok voice

uses the grok api key set in the text-models block. unfortunately the emoji-based trigger system doesn't apply to fully voice models :( i.e. changing locations / changing clothes and all other events won't fire. but image generation works fine.

a new trigger system seems doable, i'll think about it

Fiona's vrm

leveled up a bit in blender and vroid studio, but because of vroid's fundamental editor limitations, making the model i had in mind is pretty painful. vroid has a very recognizable base: the same face plasticity, weird head/body shape restrictions, hair lives by its own rules.

still, i managed to pull something out: fixed face proportions, reworked textures, livelier hair, tweaked materials and finished the model a bit in blender. Fiona no longer looks like a default dummy — at least she's starting to resemble the character from the game.

the plan is to keep refining the models, and then the animations for them.

pls write how you like it (compared to the game's other 3d characters), maybe i'm trying to squeeze something out of vroid for nothing and should switch to something else

main screen

it long bugged me that on the main screen the characters don't match the background in color and light. maybe from the outside it seems like a trifle, but in my view the start screen with the menu is the first thing that greets the player, it sets the tone — the user doesn't know anything about the game yet, but already reads its coherence.

anyway, now it looks ~ the way it was meant to, yay •ᴗ•

the only thing, i had to sacrifice the custom cursor on the main screen — because redrawing on canvas causes annoying cursor flicker, i'll probably find a solution eventually, but for now it's a compromise

✨ poker minigame

✨ redesigned the bottom settings block

✨ new sandbox images for Laura and Fiona

✨ koboldcpp: added gpu layers to settings; updated to 1.110

✨ added the option to buy the person key with crypto

✨ story:

→ Laura's arc in the prologue rewritten (if you don't go to yoga, and the variant after talking with her);

→ rework of falling asleep on the first day (Night-1);

→ reworked falling into the loop in the prologue;

→ rework of meeting Fiona

🛠 sped up the appear animation of story navigation buttons

🛠 fixed the animation of Fiona's 3d model in the story

🛠 brought error and no-network messages up to date

🛠 fixed how hints work when using the local koboldcpp model

🛠 fixed the gemini "hanging" error

🛠 fixed contexts for the english language in the story part

🛠 android: fixed settings for image generation via promptchan

poker:

opens via the phone button in the sandbox.

strip mode starts after one of the participants is out; if at the end of the game it's you and one more character left, after the win/loss it moves you into the sandbox with that character

gpu layers for koboldcpp:

should have added this setting earlier, sorry for being dumb :):

it'll help everyone who had memory overflow (especially if besides the game something else hogging VRAM was running, e.g. photoshop).

for a more stable and better result you need to change this parameter and find the balance for your hardware.

unfortunately in the latest koboldcpp versions, with auto gpu-layer selection -1, it doesn't show the number loaded, so you have to guess

person key with crypto:

for me as the creator it's the most direct and convenient way to get your support. besides the password it shows the current links in the game. comes out a bit cheaper than a regular subscription.

as for convenience — for the whole term the plan is active (3/6/12 months) you don't need to enter the password •ᴗ•

✨ person: custom contexts: can be saved locally or published

✨ new cursor replacing the "badass cursor", now the default

✨ koboldcpp: added a choice of 4 modes — auto/cuda/vulkan/cpu

✨ in some story spots, when picking a dialogue, added icons of characters you'll meet on that path

✨ event with Cas in the prologue (story navigation point "prologue, night")

✨ rework of the hints bot (quick replies): instructions changed; options became more varied, higher quality and more fitting to what's happening; they now take the protagonist's (your) system context into account

✨ new music; the name of the playing track, buttons to loop and to go to the previous track

✨ updated the cafe and kitchen locations

✨ new items in the gallery

✨ added images to the start screen

✨ Jo: added images, updated the poor ones

✨ setting for image-prompt generation — last N lines, max 3; at value 1 it takes only the user's line

🛠 ui-ux improvements; refactoring; fixed transition animations

🛠 koboldcpp: fixed launch on amd; updated to 1.109.2

🛠 reworked the shortcut from the prologue into "part 0"

🛠 improved the instruction for deciding when to roll the d20 in dnd mode

🛠 changed the trigger emoji for moving to the next day, to reduce false positives

🛠 fixed the english language in the dinner scene in the prologue

🛠 added to recommended models: claude-sonnet-4-6, grok-4-1-fast-non-reasoning, gpt-5.4, gpt-5.3-chat-latest (censored gpt)

🛠 fixed layers in chess

🛠 in the sandbox, fixed track switching — it no longer breaks the game's event queue and doesn't cause duplicate messages; it now works in "radio" mode by default — after a track ends it plays the next one

on the first publish to custom contexts you need to confirm your email by entering a code → then come up with a login and password. i hope the backend works without issues and everyone who wants to share manages it :):

the language of the published context is taken from the "ai language" setting

✨ new male character — Casper (Cas)

✨ local speech synthesis from qwen

✨ in the win and win-local builds, for convenience, added a check for the required coqui and qwen speech-synthesis libraries and download links in settings

✨ koboldcpp for images — added gpu/cpu choice; for gpu cuda has priority, if not found it launches vulkan; cpu isn't recommended but in some rare cases can be useful

✨ confetti effect at important story moments

✨ new music

🛠 fixed a critical bug with message history; by default history is now limited to 18 messages (not counting system contexts) and the slider works correctly. the max value was raised from 40 to 50 (that's a lot and far from every model can handle it, so if a local model starts to struggle or gets too expensive via api, lower it to 27-32)

🛠 koboldcpp: the image and text models no longer launch together unless that's selected in settings; set launch priority for the image model -gpu/cpu; updated to 1.108.1

🛠 added info that Ava for the sandbox unlocks as the story progresses; the unlock was moved closer to the start of the part where Ava comes to; in story navigation that's the point "in the park with Ava"

🛠 updated whisper (local speech recognition); the model was changed to the small and fast ggml-small-q5_1

🛠 improved local speech synthesis from coqui (XTTS-v2); fixed behaviour where punctuation marks were misread and glitched as 'point', pauses became more natural

🛠 re-laid out the main screen, the custom character moved into a button in the bottom-right corner (click the icon — a screen with a choice, hover and click the image — quick jump to the character)

🛠 improved snow performance

🛠 fixed detection of the current relationship with a character for models via koboldcpp

🛠 slightly increased the length for the image-prompt bot

🛠 improved error messages for text models — instead of an uninformative "crash" window, a short message

🛠 code refactoring and optimization; minor ui-ux improvements

Cas is only in the sandbox for now, he'll appear in the story later; he has much less visual content than the others so far, but got a well-developed context

qwen3-tts: creating a voice from a description is still in development; right now cloning and preset voices are available

message history: due to a bug all models except gemini were limited to a history of 12 messages, thanks to everyone who reported the problem

whisper: changed support for nvidia 50-series so as not to drag extra weight into the build, but it should still work on new gpus and fall back to cpu as a last resort; if not — pls drop a comment

✨ comfyui support

✨ rework: the very start of the game, before arriving at the house; the second day walking around the city with Jo

✨ "navigation" by arrows when walking appeared in the story

✨ the branch "if you fall asleep on a bench in the park with Ava" was merged into other branches

✨ in settings, hints got one more state — a button: instead of creating hints on every reply, it creates them on press

✨ locations: added a lecture hall/classroom; updated the city, beach, park, apartments; redrew the poor ones

🛠 https connections allowed, custom endpoints are no longer blocked; fixed the server side with password and version checks

🛠 sandbox: fixed loading of autosaves and saves — they now load correctly and don't wipe the log and message history

🛠 the "triumphant" return of copy/paste buttons for input fields

🛠 fixed values in the story editor that were applied incorrectly

🛠 fixed the position of the speech-recognition button

🛠 claude anthropic: fixed hint generation, the empty ": " message before generation no longer appears

🛠 fixed the fireplace with Fiona on android

🛠 updated the help section

🛠 koboldcpp updated to 1.107.1

🛠 minor interface tweaks

the hints button when opening the sandbox appears after the second reply

comfyui

in the in-game help section there's an instruction with screenshots, briefly the important bits:

all settings are done in comfyui; for changes to apply to requests from the game you need to update the workflow json file via export api

- in comfyui settings pick port 8188 and put "*" in the "Enable CORS header.." field

- prompts are sent from the game to comfyui, so in your workflow it's important to rename the fields to pos_prompt and neg_prompt (the game looks at fields starting with pos and neg)

- save the json via file → export (api) and select it in the game settings

example workflows are in the resources/comfyui folder:

wf-mltc-base.json - a minimal scenario

wf-mltc-face.json - with face swap by reference

i have little experience with comfyui so a more detailed guide isn't planned for now, but the tool is popular and there's lots of info; if something's unclear i'd suggest asking a neural net (chatgpt, or any other thinking model), right with screenshots or (even more convenient) — hand it the exported json

share your workflows in telegram

have a good game •ᴗ•

✨ rewrote the system contexts of all characters: the personalities got more depth, detail and individual quirks; optimized, became ~10-15% shorter

✨ koboldcpp: now for images too; if koboldcpp is selected in settings but the model isn't running, it automatically spins up the model on demand; updated to 1.105.4

✨ safe place with Fiona by the fireplace (ask to go to the fireplace)

✨ added gemini-3 models

✨ added a visual effect for connecting to Ava's chip, now picked randomly from two; + a button to turn the effect off

✨ the save button now works in spots with a text field too; on save it creates a summary (what happened over the last n messages, the message count is taken from the "message history" setting) with the character currently on screen

✨ custom endpoint

🛠 fixed a desktop bug where you couldn't click the hints

🛠 desktop: changed the area around the text field so you can scroll long answers and the invisible wall doesn't cover the 2.5d/3d

🛠 chatgpt-5: fixed the bots (who's speaking now, summary and quick replies) that relied on the old api revision

🛠 fixed remote koboldcpp tunnels for text models

koboldcpp for images:

can run alongside the text model, more flexible and cooler than stablediff cli, but:

out of the box it can only do one LoRA and applies it a bit weirdly, so when using koboldcpp only the LoRA with the highest strength is taken from settings, the rest are ignored. and kobold also caps the image size. if i manage to wire up using several LoRAs and remove the size limits, koboldcpp will become the main local tool in the game and stablediff will retire so as not to drag extra gigabytes into the build.

remote (ngrok/cloudflare) for images will be added later

for the standard ilustmix model (and many others) you need to specify the vae file resources/stablediff/sdxl.vae.safetensors in the game settings

summary via the save button:

story and sandbox summaries are separated. for now it only works with the "message history" setting enabled, needs extra testing; if no issues come up it'll become available in the next release, for everyone without a password, and on by default •ᴗ•

custom endpoint:

https only, still needs testing; possibly due to cors security policies it won't work for you — write which provider didn't work and i'll add it to the exceptions in the next release

✨ image generation via api (uncensored): added support for the promptchan service

✨ story, in the prologue: changed Fiona's visual in the shower; a small variation in the opening conversation with Fiona; rework of the branch with Dina at night (if you nap for a couple of hours and go to the kitchen). updated Ava's story images. rewrote Dina's system context (closer to the sandbox one)

✨ sandbox: updated Ava's and Laura's images; updated the warm clothing for Fin, Jo and Dina

✨ sound on a successful image generation

✨ new items in the gallery

🛠 common settings for image generation moved into a separate block

🛠 story navigation in a collapsible block

🛠 for local image generation, max sizes increased to 2048x1536

🛠 windows: fixed 3d vrm

🛠 changed the links to boosty instead of patreon (what jerks they are, rrr)

🛠 minor ui-ux tweaks

🛠 added an instruction with promptchan links to the help section

🛠 in win-local default LoRAs: hyperdetailed illustration and sexy details

promptchan: before the game had a local model i used them for the explicit visuals; by default the style Illustration XL+ with the Artsy 2 filter is set, the hyper filter is also good.

tokens without a subscription give the regular styles; XL styles require a subscription to their not-cheapest service, but if you cancel the subscription they'll, in the finest adobe tradition, offer a huge discount for a month.

on android you have to go via a proxy because of CORS and preflight checks; it should work like on desktop, maybe a touch slower.

if you can recommend some other thing for uncensored api image generation — pls write, i'll try to add it

✨ story, branches: morning at Josephine's apartment (if you stay overnight with her together with Ava); morning at Fiona's

✨ in the log: added buttons to check the current relationship with characters, and animated images; an editor for story choices

✨ new trigger when the relationship level with a character changes

✨ kitchen location

🛠 for 2d, the image now dims only while the text input is focused; for 2.5d/3d fixed the field's look during multiline input

🛠 android: for the story part, fixed the rooms block when moving around the apartment; fixed the copy/paste text menu

🛠 quests/tasks: if a reward is received the task no longer hangs in the list; a small redesign

🛠 redrew poor character images and backgrounds for the winter start screen

🛠 replaced the vrm models in the story part

🛠 optimized the location-detection bot (made it shorter)

🛠 image generation in the story now works correctly with narrator text

🛠 changed the trigger-firing scheme: now if there are several in one message they all fire, not just the first

🛠 in nvl mode fixed: button display and the text field; a bug where images weren't generated

🛠 minor interface tweaks

🛠 fixed speech synthesis from elevenlabs

🛠 koboldcpp updated to 1.103

🛠 fixed a 2.5d menu bug when using koboldcpp

🛠 win-local: default LoRA updated to cryosistyle v3

if you're here for the story continuation — i recommend waiting, the bits being written are short for now and will still change

✨ 2.5d: the angle changes when moving the mouse / tapping the screen, zoom by scroll / pinch gestures; added a render-quality slider to settings, default 0.9

✨ 3d: each character got their own vrm model instead of the crappy readyplayer ones + undressed models

✨ added one more state that clears the emoji triggers from the game. "remove sexy context and personality" — now also removes the short character info, leaving only the gameplay triggers

🛠 added to recommended models gpt-5.1-chat-latest, gpt-5.1 and claude-sonnet-4-5-20250929 (still the most 'human' model, top1)

🛠 if a location is changed from the menu and it's the same type as the current one, the bot isn't called

🛠 fixed chatgpt working via the extended model selection

🛠 minor ui/ux tweaks

🛠 redrew some poor images, especially faces; optimized the in-game backgrounds, redrew the poor ones

🛠 when using 2.5d for animated backgrounds and characters, made alternative static images with more detail

🛠 fixed the display of tasks when hovering over notes

🛠 improved the performance of Ava's chip visualization; now if the image is generated the visualization isn't shown

3d: the models for Fin (couldn't find a dick, so he's in shorts) and Ava are rough, the rest are better but still don't reach the quality i want, though already closer; i'll keep developing and periodically updating the models; if you want to try your hand in vroid studio — all the original .vroid files are here — use them, change them, edit them, create your own

2.5d: there's room to improve, i haven't yet managed to get rid of the jaggies and i'm sure you can make a more obvious parallax for distant background objects

and a few more details

2.5d and 3d don't work with two characters yet

for 2.5d on android the quality is deliberately reduced by 0.2 from the set value so it doesn't lag

✨ a character's thoughts in a separate animated block (i think it's livelier and more aesthetic this way)

✨ a tasks/quests system

✨ 3 settings moved into experimental, and now available starting from the base tier

✨ windows, linux, mac: added loadable lists of all available api models for gemini, openrouter, openai, anthropic, grok, cohere, groq; the 'recommended models' toggle has retired

✨ all characters became less verbose, but will be wordy if asked or if circumstances require it

✨ added items to the gallery

🛠 in the dialogue history, fixed a bug with 'tearing' the narrator's text in streaming mode

🛠 improved the censorship bypass for api models when forming the prompt for local image generation

🛠 fixed an english-language bug for quick replies

🛠 redesigned the 'erase' functionality for part of a character's context, added color marking, different wording, and an explanation in the help section

🛠 fixed a claude bug when two characters are on screen and streaming mode is on

🛠 fixed the story animation introducing the characters (bottom to top)

🛠 default number of messages in history increased to 18

🛠 number of messages in tamagotchi memory reduced from (45, 90, 135, 180, 225, 270) to (27, 45, 90, 135), 45 recommended; higher values would be too expensive for api models, and local ones won't handle it

🛠 increased the scrollbar width in locations and the log

🛠 optimized gallery images/videos and character images

🛠 koboldcpp updated to 1.100.1

🛠 fixed the nvidia/amd switch display on entering the game

🛠 thoughts hide on click, and reappear on a repeat click

🛠 in win-local changed the base text model saiga_nemo_12b to a smaller quantization Q4_K_M (didn't notice quality drop much, and it works a bit faster)

🛠 optimized images, videos, styles (scss)

🛠 fixed a koboldcpp bug with streaming output

🛠 fixed the interface, minor ui/ux tweaks

🛠 added a setting "don't limit ai answer length" that disables the instruction limiting the ai's verbosity

✨ reworked the help section (the friendliest version for new players)

✨ image generation:

- optimized the negative prompt (turns out one that's too long can lead to less crisp details)

- improved automatic prompt writing: now if there are no people in the scene the bot focuses on something else, especially useful in dnd modes; got a bit better with situations where 3 characters are in frame

- on generation it takes the last 3 messages into account instead of two

- updated the default model to ilustmix_v9 (quantized to q8 — smaller size, no visible quality loss)

- fixed automatic image generation via api for grok and openai (the every-n-messages setting is taken from local generation). for openai added size

✨ tamagotchi

- now if the cursor hasn't moved for more than 5 sec, Mika stops watching

- reacts if you stare at her chest (5 clicks)

- Mika learned to generate images, uses the sd settings from the main game, but the size is fixed at 768x1024

✨ in the main game updated the 3d for Fiona (the model from tamagotchi)

✨ in the story part rewrote the bit with Fiona changing clothes, now it's vrm 3d there

✨ added an airport location

✨ android: added loadable lists of all available api models for gemini, openrouter, openai, anthropic, grok, cohere, groq. the 'recommended models' toggle has retired

🛠 in the story part slightly polished the very start of the game: added airport sounds, images, improved the quality of old images, added a new kind of animated text

🛠 improved gemini error handling; instead of a stub it now returns the real errors from the model

🛠 reworked the look of info windows; added a message when there's no active text model and a message to set up speech when 3d is enabled in the story (shown once)

🛠 fixed the link in settings for testing the openai.fm voice

🛠 a different icon for the koboldcpp console (it was indecently bold before)

🛠 android: fixed selection and the caret for the multiline text field

🛠 koboldcpp updated to 1.99.4

🛠 re-laid out the local kobold models block in settings

🛠 local speech synthesis: fixed the language/accent, it's now correctly taken from the 'language ai' setting

✨ tamagotchi (windows)

→ Mika (the agent) learned to set timers, just say/type: Mika, set a timer for n minutes/seconds/hours

→ follows the cursor

→ fixed panel hiding and the drag button

→ outfits: Fiona — red, club, strawberry, mesh dresses; frog dress; skirt; stockings; shorts and top; in latex; no clothes. and a sexy bunny

→ added breast physics (not perfect yet, but god is it fun to make)

→ pdf instructions

✨ chibi anger animation for Dina when she's mad

✨ branch with Dina and Ava

✨ story navigation by rooms of a location (Dina's apartment)

✨ browser speech recognition moved to a local server (i was dumb and didn't know you could do that)

🛠 save fix

🛠 koboldcpp updated to 1.98.1

🛠 code optimized and refactored

🛠 fixed font selection

a small story branch in the morning at Dina's; if you want a coherent continuation — skip it and come back when the story update is ready •⩊•

→ story mode now supports image generation - prompts are now built to match the current location and characters on screen

→ new pool location

→ redesigned location screen with captions

→ added claude-opus-4.1 (noticeably different from claude-sonnet-3.7, but costs way too much - 5× more, use with caution)

→ support for gpt-5, gpt-5-chat, gpt-5-mini, gpt-5-nano

→ sd:

- improved prompt generation for images

- sd server (hot model) now auto-restarts up to 3 times

- sd doesn't like long prompts - for auto-generation they're now forcibly trimmed to 540 chars to avoid crashes

- max length for negative/user prompts limited to 270 and 180 chars

- settings now have a "self description" field

- fields for manually editing clothing and location

- fixed a bug where the default LoRA was applied at game start, requiring opening settings to load the chosen ones

→ local speech synthesis chunk size increased from 200 to 450 chars

→ koboldcpp updated to 1.97.4

→ fixed openrouter model switching - no restart needed when changing from the popup

→ added cuda support in sd for linux (except 50x cards), fixed img_gen error, appImage size greatly reduced, resources for koboldcpp, stablediff and whisper moved into a separate folder

→ fixed "invalid mode txt2img" error for mac

gpt-5:

temperature doesn't work for new models - it's stuck at 1, and top_p (which was 0.9 in-game) no longer has any effect. censorship hits hard on the "thinking" models (gpt-5, gpt-5-mini, gpt-5-nano), but can be partially bypassed by continuing sexualized plots with them after starting on gemini / claude-3.7 / gpt-4.1.

gpt-5-chat is better - censorship is about the same as gpt-4.1, maybe slightly more vanilla

sd: for better accuracy, raise cfg-scale to 8-9 and use extra LoRAs sparingly - only ones that affect overall style. LoRAs with baked-in poses/characters reduce relevance

✨ windows: local speech synthesis

✨ local image generation: → added launch via a local server so the model isn't unloaded from memory; how stable it is isn't clear yet, so both variants are kept → test support for flux models

✨ spectrum analyzer for openai, google and local speech

✨ updated the gallery

✨ new music

✨ nvidia/amd switch (instead of automatic detection)

✨ desktop: a small 'easter egg' on the main screen (5 clicks on the game version)

✨ koboldcpp updated to version 1.96.2

🛠 the empty input field now hides after 3 seconds

🛠 fixed system contexts so characters should become a touch less horny (just a touch)

🛠 brought the help section up to date

🛠 fixed comics mode

🛠 openrouter: fixed image-prompt generation

some quirks of local speech synthesis:

→ on first launch the annoying firewall will ask for permission. model warm-up is 20-50 sec, then much faster, especially on gpu

→ put the .wav voice reference into resources/speech/, it won't work from other folders

→ when changing voice the model reloads

→ since the model was trained on short clips, the text has to be cut into fragments, because of which there can be pauses in long text, but on gpu they're almost unnoticeable

→ into the lite windows build without local models i didn't include the binaries since they're quite heavy (building them was a pain, it most likely pulls extra libraries, but i haven't managed to make it "slim down" yet)

→ higher quality - to my taste, if you use your own wav, especially non-english; the language is taken from the "ai language" settings

→ for mac and linux it's planned

i couldn't really launch and play with the flux models, but maybe you'll manage. 8gb of vram is little. i didn't include it in the general build; if you want pain — try it, the archive has quantized ones (maybe that's the problem), but you can replace with your own clip_l and t5xxl, the game looks for them in resources\stablediff\flux, put the flux model into resources\stablediff\models, the game filters by name so the model name must start with flux

to everyone who's more about the main story: i get that i played a bit too much with tech, i'll get to the story part in one-two updates, and i think by mid-september there'll be a continuation. have a nice week •ᴗ•

✨ stable diffusion:

- changed the folder structure, files are now spread across the cuda, darwin, vulkan folders, models in the models folder

- the default model converted to gguf became faster by 1-2 seconds on gpu and i think by 10% on cpu, build 2.7gb smaller (yay)

- added support for amd cards, the game itself detects the card and hooks up vulkan instead of cuda

- improved location detection and memory

- changed the LoRAs in the win-local build

- rebuilt for mac and amd cards, increased the graph sizes

✨ koboldcpp:

- updated to version 1.95.1

- koboldcpp remote: added ngrok support, as an alternative to cloudflare

- fixed the bots: for detecting clothing change, location; the image-prompt bot

✨ added the Rounded Mplus 1c font, now the default

✨ added a photo-studio location

✨ desktop: added a button to hide the character and text, hotkey shift+H

✨ added the grok 4 model (pretty unstable for now, but interesting in places)

🛠 Laura: tweaked the system prompt for image generation

🛠 changed the input field, typing large multiline text became more convenient

🛠 tweaked the adaptive font size, the text will now fit without scrolling much more often

🛠 brought the help section up to date

🛠 in the quick menu only icons are left

🛠 changed the icons for saving an image and for the phone

🛠 settings: tabs now collapse on a click across the whole block, not just the icon

🛠 fixed the button for saving a generated image in nvl mode

🛠 rewrote the location-detection bot, it now launches right after receiving the trigger instead of on the next "turn"

🛠 removed the note on the main screen about recommending gemini

🛠 hints (quick replies): now the last 4 lines are used to form them, not the whole dialogue

🛠 openrouter: fixed an error with selecting models from the dropdown

🛠 cosmos: added the cosmosrp-3.0 and cosmosrp-3.5 models, deprecated the outdated ones

🛠 anthropic claude: fixed the empty ": " message error when the small bots run; hints now work

🛠 cohere: added hints support

→ ngrok plugin (instructions in the help section, koboldcpp-remote)

→ for weak machines i quantized a more compressed q5 model, should be a bit snappier

→ rebuilt sd supporting 50-series cards, put it into the resources\stablediff\cuda folder

→ local stable diffusion and LoRA models

→ button for saving images

→ added models: gemini-2.5-pro and gemini-2.5-flash, gemini-1.5-flash-002 removed

→ koboldcpp updated to version 1.94.2

→ person: a prompt field for the local image model has been added to custom characters

→ android: fixed fiona export to telegram bot

→ fixed a weird issue with gemini models where the last user message was duplicated

→ game commands like "we moved to..., react appropriately" are now aesthetically hidden

patch for amd gpu - unpack the archive into …\resources\stablediff\ and overwrite existing files - required for local image generation

the main points have been included in the in-game help section;

below are the things I forgot to mention:

→ for the default iLustMix model, use clip-skip = 2. if u get a black image instead of a picture, it means your selected model needs a different clip-skip value - try setting it to 1 or 2. I haven't seen value 3 used, but I left it just in case

→ custom character settings now have an appearance field

(context for text models will be added later, there are some structure issues)

→ write custom character prompts and those from the quick generation field (button in the bottom right) strictly in english, stable diffusion doesn't work with other languages

→ the clothing detection bot uses the same triggers already present in the game

→ the image prompt bot only looks at the last two messages from u and the character

→ on mac, there might be issues with image generation

if u're playing on a mac, pls take part in this open poll (unfortunately, I can't test it on a real device myself)

✨ added all the remaining openrouter models

✨ Fiona got a cat-girl costume, updated the poor images, rewrote and optimized the system contexts

✨ person: a telegram bot for Fiona (desktop)

✨ a few new images for Jo

✨ gallery: added thumbnails so it doesn't lag on weak devices, fixed and optimized the unlock function

✨ many settings items became collapsible, it's tidier and i hope this minimalism scares new users less :):

✨ win-local: the default local model changed to saiga_nemo_12b.Q5_K_M

🛠 linux: the game now looks for gguf models in the resources/koboldcpp folder, ~ like on windows

🛠 android: fixed the 6-sided die animation

🛠 koboldcpp updated to version 1.93.2

🛠 brought the help section up to date

🛠 changed the hover sounds for menu items and characters on the main screen

🛠 fixed and optimized the start-screen animation, fixed interface elements

🛠 changed the scene-transition animation in the story (i really didn't like the fade from black, so now there's even more blur there), smooth transitions for video inserts

telegram bot (beta):

desktop only and for person subscribers for now; on android the json isn't saved, i'll fix it, it'll come later

in Fiona's custom contexts, the "export to telegram" button creates a json file you need to send to the bot that opens in telegram

security: uploaded json profiles and the bot history are encrypted before writing using the Fernet scheme (AES-128 + HMAC-SHA-256)

the upshot should be that even if someone steals the json files from the server, without the key they'll only see base64 mush, not your secret tokens

the plan later is to link the bot's history in telegram with the game, so you can seamlessly move from the bot into the game and back while keeping the chat history

if it turns out fun, it'll of course appear for other characters too

saiga nemo:

i'll keep changing the model until we find the best one •ᴗ• (in my experience and from subscriber feedback this is the top one for now)

truly multilingual — it understands and writes spanish, russian, german, french, italian at an everyday level. english ~ on the level of gemma and deepcogito

✨ added openrouter support and free models: qwen/qwen3-30b-a3b:free, tngtech/deepseek-r1t-chimera:free, meta-llama/llama-4-scout:free, meta-llama/llama-4-maverick:free, deepseek/deepseek-chat:free, deepseek/deepseek-chat-v3-0324:free)

✨ log: editing and deleting lines, a button to quickly scroll down; on opening the log the latest lines are now always visible

✨ Dina: new images, changed starting locations, you can now go down on her, grope her chest, + blowjob images

✨ added models: claude-sonnet-4, claude-opus-4, claude-3-5-haiku-20241022, command-r7b-12-2024, gpt-4.1-mini. deprecated gpt-4-0613, gpt-4o-mini and gpt-3.5-turbo (the 16k version stayed)

✨ a local model for speech recognition (windows, linux)

✨ the speech-recognition setting is now saved, on game start it launches the chosen solution (browser/local)

✨ internal ru proxy: added the lv (Latvia) proxy server

✨ the default local model became deepcogito_cogito-v1-preview-qwen-14B-Q4_K_M, changed the list of recommended models

🛠 fixed location behaviour, now it's not the first one set but a random one from a previously saved set

🛠 fixed ui-ux animations and minor interface glitches

🛠 koboldcpp updated to version 1.92.1

🛠 now for windows, if you have a card with cuda 12 support, you can drop koboldcpp_cu12.exe into the resources/koboldcpp folder (it'll work a bit faster), the game will use it instead of the standard koboldcpp.exe

🛠 the context-trim setting for local models, default raised to 9900 characters

log: you can now edit the dialogue history, both the character's lines and your own — this can help fight the censorship of big api models and also just so the history is the way you want it

openrouter is without streaming output for now, since not all of their models support it and the syntax is a bit different, i'll add it later.

internal lv proxy: unfortunately for gemini this didn't help, since in Latvia it turned out to be blocked too, but for chatgpt and claude it works great. so for gemini the internal proxy works only via server 2 (sf); if it's blocked for you — the only option for now is your own vpn

local speech recognition: uses the model ggml-large-v3-turbo-q5_0; it works somewhat differently from the browser one, it can't stream (but that's solvable, i'll refine it), it waits until the recording finishes, so without a gpu it can be slow with a big delay; to make it faster you can try replacing the local model file in the resources folder with a simpler model, e.g. ggml-tiny-q8_0.bin, just rename it to ggml-large-v3-turbo-q5_0.bin so the game picks it up

→ spaceship location

→ a space-sexual girl outfit for Jo, Dina, Fiona and Laura; a space-guy outfit for Finn

→ person: dnd scenario - space odyssey

→ koboldcpp updated to version 1.89

→ fixed a bug where you could add a duplicate of a character already on screen

→ reworked the model-selection block in settings - it's now more compact and a bit clearer

→ ru-lang proxy server: the second kz server for anthropic models now works correctly

✨ added api models: grok-3, gpt-4.1 and command-a-03-2025 from cohere

✨ added small chibi animations for Fiona's surprise and irritation

✨ new sandbox locations: couch, bedroom, beach house

✨ story: expansion of the Dina + Ava branch if you stay to sleep at Dina's; finished the branches — if you stay to sleep alone

✨ sandbox: Jo — a few new images, updated the poor ones

✨ characters now always understand which location they're in, the last used location is remembered and loaded into the system context. the favorite location can now be turned off by clicking the icon again (if history saving is on — i recommend turning the favorite location off)

✨ added police station and prison locations

🛠 nvl mode: fixed text display for claude

🛠 fixed critical bugs with injection-instructions in the story part for the english language and the female protagonist.

because the injections didn't fire, the story narration for english was clearly full of holes, and in some spots wouldn't let you into certain branches at all

🛠 koboldcpp updated to version 1.88

🛠 in the story part fixed the phone close button

🛠 cohere: fixed the bots for in-game triggers

🛠 tweaked the detection of whether to put on warm winter clothes - should reduce false positives

✨ new images in the gallery

✨ music control in the sandbox

✨ story: a small scene in the park with Ava (if you find the bench)

✨ person: history saving in the sandbox

🛠 android: fixed a koboldcpp-remote error

🛠 android: fixed a settings-import error

🛠 koboldcpp updated to version 1.87.2

🛠 fixed the long-standing pain with input windows that under certain circumstances could overlap each other

🛠 clearing history now also clears the message log of the selected character

🛠 optimized the instructions for the summary (every n-messages)

✨ added image generation from grok (chosen in settings)

✨ story: added a branch with the police station (on leaving the park)

✨ when generating an image, the character's picture now becomes smaller and shifts to the right

✨ images for Ava

🛠 fixed an error when a local model is selected

🛠 added correct output of errors/notifications during image generation

✨ animation on the start screen

✨ micro-animation and clock sound when some time has passed

✨ a few images for Jo

✨ added the gemini-2.0-flash-lite model, deprecated gemini-1.5-pro-exp-0827, gemini-1.5-flash-latest and gemini-1.5-pro-latest

🛠 koboldcpp version updated to 1.86 (supports gemma3)

🛠 fixed an error where closing the game and reopening it within 5 seconds lost saves and settings.

when using local models via koboldcpp it's still recommended to wait 2-3 sec before reopening the game

🛠 when importing a settings and saves file, added info that a game restart is required

🛠 fixed the music in the story part

🛠 changing the ai language no longer requires a game restart

🛠 the sandbox setting for the character to speak first no longer requires a game restart

✨ a new context field for describing the world and instructions

✨ micro animations and sounds during a kiss

✨ a few new park locations

✨ added the claude-3-7 model

✨ added the gpt-4.5-preview model

✨ updated the lore and characters section, added statistics

🛠 apk: fixed claude working

🛠 optimized characters' system contexts in the sandbox and images in some scenes

🛠 Laura is now 28, and Fin is 29

🛠 fixed the kz server for the internal proxy, added streaming output for claude with the proxy

statistics: for now it just counts how many times the character and the user came -`♡´- it'll be refined so it can be used in the system context

→ gallery: added a multitude of images, now to open them you need to use ⦵, which can be farmed by winning mini-games: chess, save the princess and collecting macrodata

→ Jo, Laura, Fiona and Ava: now they can go anal (trigger and images)

→ Laura and Jo masturbate (trigger and images)

→ a gallery option has been added to the quick menu

→ mini-game with macrodata (the very same ones collected by the workers from the show - severance), available from the phone or by asking Ava

→ added model gemini-2.0-flash

→ groq: added model llama-3.3-70b-versatile, deprecated llama-3.2-90b-vision-preview

→ added park locations

→ fixed bug with english context when adding a second character

→ koboldcpp updated to version 1.84.2

→ removed the button for activating 3d from the story section

→ apk: fixed voice selection for elevenlabs

→ fixed bug with the car, when the location did not change after the car

→ person and base passwords changed

for "save the princess" the most points are given - 40

for chess (losing is almost impossible) - 10

macrodata - random from 1 to 18

macrodata: the idea was spotted and swiped from the expanses of codepen

for everyone who hasn't watched the series (severance) - I highly recommend it

→ speech synthesis from elevenlabs (also on apk)

→ sandbox: demo of 3d characters with lipsync

→ when first opening the game, the language selection for ai is now set correctly

→ selection of cosmosrp-pro and chatgpt-4/4o models is open and no longer requires a game subscription (without person password)

→ updated list of starting models on the welcome screen and in settings (if there is no local model in the game folder)

3d:

one of those ideas that came almost immediately but was put on the back burner my knowledge in 3d is quite superficial and not enough to create a model myself, so I used the free very limited editor ready player me, models are boring without any hint of sexuality, but together with lipsync demonstrate what I want to do

plan: find a suitable 3d artist on the marketplace and order a character based on refs from the game (later we will decide who will be first)

approximate level of the model that I want

if u know something that can help - write ) the model should be compatible with ready player me + arkit, oculus blend shapes

with the 3d avatar, only speech synthesis from google and elevenlabs works so far,

fps locked at 45 frames

speech synthesis from elevenlabs in my brief view is top: a large selection of voices, trial, slightly cheaper than openai, and no worse in terms of level

the continuation of the story is still in development. unfortunately, plans have shifted somewhat (there were many important bug fixes). the new deadline for continuing the plot is the end of february to the beginning of march

✨ sandbox: when changing location from the menu, characters comment on the move; you can set a favorite/starting location for each

✨ added the bypass_eos setting

🛠 koboldcpp version updated to 1.82.4

🛠 updated the help section (hopefully friendlier for new users)

🛠 fixed autosaving; the slider now also affects automatic dialogue trimming, the recommended value is set to 18, the max reduced to 40

🛠 fixed steam activation in the shower when chosen from the locations menu

🛠 fixed the ai-language selection in the dark theme

🛠 the winter theme is extended until february 13

🛠 fix for the save-the-princess minigame (my clumsy hands didn't close a bracket in the code, dammit)

🛠 the sandbox setting with generated answer options now works correctly (it was broken and worked backwards, heh)

🛠 story: now if at the end of a short series dedicated to conversation a character asks a question, the answer isn't replaced with an ellipsis but is shown in full with the note "the question hung in the air"

🛠 koboldcpp logs are now always trimmed to the last 360 lines

🛠 on opening the sandbox the "previously happened" reminder button now correctly appears only if this character's history was saved

🛠 fixed the order of save/load slots

🛠 fix for when the dnd dice fly off the table, there's now an option to reroll

bypass_eos - use it if a local model outputs "EOS token triggered" too often and out of place

autosaving had a bug — a hardcoded value of 20, which meant a larger value in settings (now affects trimming too) was simply ignored and the save didn't fire

✨ a phone-calls mechanic in the sandbox. it's being made for the story part, in the sandbox for testing :) a bit toward immersion

✨ added a language choice for the ai (Français, Español, Português, Deutsch, Italiano, Українська, Slovenský, العربية, Türkçe, 한국어, ภาษาไทย, Polski, 日本語, Tiếng Việt, ქართული, Հայերեն)

✨ release on mac, yay you mac junkies!

🛠 fixed the die for dnd modes to a d20, added sound, fixes to the buttons' look

🛠 now if a local model is selected, the game doesn't require a network

🛠 fixed the version and password check calls when there's no network; in some situations, especially on a phone, they could cause the app to hang

ai language: affects only the chatbots' text. since the system context is written in english, characters can sometimes slip into english, but in the sandbox this happens much less often and they're easier to bring back to the chosen language.

the phone mechanic is new, it may be glitchy on some models

✨ integrated koboldcpp with a local model*

✨ desktop: added a welcome screen with a choice of a local model or api

✨ in the sandbox added a location choice (top-left corner)

✨ now if koboldcpp and koboldcpp-remote are selected, the kobold tab in settings is open by default

✨ updated Ava's poor images

✨ redesigned the choice buttons

✨ reworked the path to one of the sexual scenes with Dina in the story

🛠 minimum screen size in windowed mode increased from 640x360 to 900x500

🛠 improved the bots detecting location changes and the arrival of a second character. fixed working on local models via koboldcpp

🛠 max temperature value lowered to 1.4, recommended to 0.72

🛠 koboldcpp:

→ now on closing the game the koboldcpp and electron processes terminate correctly and don't stay in the task manager

→ fix for early token-generation termination (EOS token triggered! ID:2 error)

→ adjusted the settings

→ fixed hints and generated options working

→ max answer length reduced to 630 tokens, recommended value to 190

→ a koboldcpp shutdown button appeared

→ added info about approximate system requirements on the start screen and in settings (if there are no models in the folder)

→ added custom stop triggers

🛠 fixed a version-check bug

🛠 in the story fixed the earthquake animation (and possibly some others)

🛠 replaced the music in the story event with Fiona in the shower

*now for windows there will be three variants: standard (together with koboldcpp), -local (with a local model) and -lite (without koboldcpp); for linux you'll need to download the model separately and select it in settings (i'll sort out the configuration later and do it the same as on windows with the resources/koboldcpp folder

windows: koboldcpp with models is in the resources/koboldcpp folder, the game will load all gguf models from this folder onto the start screen and the settings tab

windows/linux: a model from any folder can be loaded from the settings tab by clicking "select model"

the local model i settled on — Gemma-2-Ataraxy-v4d-9B.i1-Q4_K_M — feels ~ like chatgpt-3.5 level, weak, but much better than cosmosrp at handling emoji-triggers; i think it should work fine for most users

of the models i tested and which might suit the game more but is more demanding — ChatWaifu_v1.4-Q4_K_M-GGUF (if there are no models in the resources/koboldcpp folder, it'll show the system requirements and a link on the main screen and in settings)

✨ the spirit of christmas and new year

✨ added grok-2 models

✨ now Dina, Jo, Finn, Laura and Ellie can dress warmly if it's cold

✨ added winter city and forest locations

✨ added functionality to transfer game settings and data (bottom-right corner)

✨ Fiona: updated the poor images, added new ones

✨ added snow in locations where it can fall

✨ for optimization, on the load and save screens added a setting to show images, now hidden by default

🛠 reworked the interface animations

🛠 fixed the order of autosave slots

🛠 fixed the dark theme

🛠 fixed bugs with {{userMessage}} and empty messages (but that's not certain)

data/settings transfer — a feature needed in the future, and also for those who just want to move from one device to another. android users will need it a bit later once i fix the wrong app identifier value

showing images on the load/save screens — the problem is that with many slots on weak machines it can significantly affect performance, because it uses the in-game images, not thumbnails, especially if they're animated backgrounds. i'll bother with creating thumbnails later, for now it's like this

✨ now all sandbox characters know what day it is today

✨ added a user birthday choice at the start of a new game (can be changed in settings)

✨ added a setting to disable generated dialogue options in the sandbox

✨ person: added D&D scenarios to the sandbox: Cyberpunk 203X and zombie apocalypse

🛠 threesome: fixed detection of the character being added when called from chat

✨ added a feature with two characters on screen to the sandbox

✨ support for the grok-beta model (the one from Musk)

✨ a button to force Dina back onto the screen

🛠 fixed a critical error with custom contexts for the female protagonist, now everything works and dnd mode too

🛠 max temperature reduced to 1.8

🛠 removed the gemini-1.0-pro series models and gpt-3.5-turbo-1106

🛠 fixed the dark theme

🛠 apk: less screen dimming on text input

🛠 gemini: optimized the bots responsible for location changes and saving history; the location bot now works correctly, always creates a new empty session and looks only at the last 2 messages instead of the whole history; the history-saving bot no longer has excessive duplicated text. an especially relevant fix for long game sessions

🛠 in the prologue, removed the shameful (low-quality) videos with Dina and Laura from the dream, replaced with images

🛠 music files re-encoded to ogg (same quality but weighs half as much)

🛠 all webm videos compressed a bit with no visible quality loss

beta

🛠 optimized/removed/replaced poor animations and images

🛠 fixed a steam bug where a repeat trigger activation added another steam, which could significantly lower performance

✨ koboldcpp: added remote mode support; to solve the problem of characters starting to speak for the user, added a field to exclude the name form — relevant for languages other than english

✨ api keys and the password are now caringly hidden

✨ Fin: 2 new images

✨ redesigned the settings page

✨ updated the help section

🛠 fixed the mechanic for choosing dialogue/action options, it should become a bit more adequate; the number of options increased by one

🛠 apk: fixed a bug where the dark transparent backdrop on collapsing the keyboard didn't cover the scene fully

🛠 redrew poor images in the hospital

✨ story part:

→ a new mechanic with generated dialogue/action options*

→ location changes now work in the story part too (it brought more freedom, but maybe a bit more chaos)

→ redesigned the popup windows

→ redrew poor images of Dina, Fiona and Laura

→ redesigned the prologue event at the table, fixed errors during Laura's conversation with Dina

→ a new event for Laura in the bathroom, in the prologue with the doors.

→ person, story navigation: became available not after finishing, but right away; added prologue points (dinner and doors); fixed a bug where pressing a navigation point stopped the chatbots from working

→ apk: fixed the keyhole display

✨ a few new sandbox images for Laura

🛠 optimized the background animations, on weak machines it should get a bit better

* works ~ like with hints, but: without the ability to write your own text, from 4 to 6 options, with attempts at humor and weirdness. it's also added to the sandbox, but with a very small chance of appearing <1% (more precisely, every 16th message has an 18% chance)

✨ Ava: added new images, and starting locations; when connecting to the user's chip — a slick scripted background

✨ gemini: added the gemini-1.5-pro-002 and gemini-1.5-flash-002 models

✨ 4 new tracks in the sandbox

✨ info about a new version is now shown on the main screen (if the version is below the current one)

✨ added a keyboard-typing sound (used in the story part)

🛠 gemini: fixed a bug where text entered on a failed request wasn't added back into the field

🛠 story part: fixed incorrect display of pronouns and forms of address for the female player character; added an avatar of the female protagonist in dialogues; fixed errors (in some places the narrator wasn't loaded); fixed transitions and errors in dialogues

🛠 slightly optimized the initial game load

✨ new locations: stadium, green field, abandoned factory/warehouse, mage's tower/ancient observatory, ruins of a sky city, crystal cave, icy plateau in the mountains, bdsm room, rooftop

✨ Dina: redrew poor images, added new ones; fixed Dina's return in case she left

✨ Jo: redrew poor images

✨ kobold is still in beta but became available to everyone without a subscription

✨ DnD mode: rewrote the system context for dnd mode; added a slick 20-sided die, now used on all rolls

✨ person: saves in the sandbox

🛠 reworked the auto-save-history-every-n-messages functionality. it's now a toggleable option. off by default*

🛠 on the load/save screen the slots are now shown left to right

🛠 start-screen tweaks

🛠 fixed errors related to incorrect display of the "where we left off" button and clearing history when part of the context is removed

✨ koboldcpp: added streaming output, fixed system-context passing and chat-mode operation

✨ you can play chess with Dina*

✨ temperature (freedom level): max value raised to 2, recommended to 0.81. values above 1.0 for quirkiness and more creativity

✨ windows, linux: a new cursor was added, less demanding on resources, should run smoothly on weak machines too (if there are no complaints about it — i'll make it the default on first open)

✨ sounds: dice tossing, a shattering glass jar

🛠 ux-ui tweaks

🛠 on phones a small redesign of the start screen

🛠 apk: fixed a hint-overlap bug

*for now Dina doesn't comment on the course of the game, she just reacts to a win, loss or draw. and it's very hard to beat her :) if the idea catches on — i'll add controls and Dina commenting on the game so she plays herself (right now a pretty dumb algorithm does it).

✨ support for local models via koboldcpp

✨ Fiona: updated poor images, added new ones, now if it's cold she'll put on a hoodie

✨ Laura and Ellie learned to undress to a top

✨ added a winter-cabin location

🛠 optimized the base contexts of all sandbox characters

🛠 updated the shower location

🛠 fixed a bug where the shower steam stayed in other locations

🛠 changed the ending trigger so it doesn't conflict with the bathroom/shower location

🛠 minor interface and help-section tweaks

🛠 save the princess: the minigame was a bit optimized and refactored; the jump became 20% higher; apk: fixed weapon selection and added a jump*

*did it become easier to beat the minigame on a phone? - yes, but it can still be hard. the-floor-is-lava and careless handling of the new weapon can only make things worse

✨ support for local models via koboldai, in beta status

✨ added models from cohere*

✨ a filter for recommended models

✨ added the claude-3-5-sonnet-20241022 and llama-3.2-90b-vision-preview models

✨ ru: the server choice for the internal openai and anthropic proxy moved into settings**

*at first glance, gemini-flash level - not very powerful, fast, not expensive. they give a free trial, which is nice

**the kazakhstan server suddenly stopped working, more precisely openai and anthropic stopped working in Kazakhstan, if i understand correctly. while i figure out where to move, i decided to route to the proxy where gemini runs; there the server works single-threaded, so it may be loaded heavier than usual for a while

✨ added animation and sound when switching character images

✨ new music on the main screen and in the sandboxes*

✨ nvl text-display mode

✨ dark mode, added a dark theme

✨ the text input field is now multiline

✨ font choice in settings

🛠 fixed streaming output for gemini (without the internal proxy)

🛠 fixed the input field

🛠 apk: fixed dialogue output in adv mode

🛠 fixed speech synthesis from google

🛠 fixed the minimum adaptive-font size

🛠 in the sandbox removed the dialogue-skip button as unnecessary

🛠 you can now select and copy text from the log

🛠 fixed the character count in custom contexts

*for now the god of randomness rules the music — random for the start-screen track and random in the sandbox. later there'll be playlists and track selection

✨ Laura: updated images, personality, history

✨ text is now shown with paragraphs, yay!*

✨ now in the sandbox, on activating undressing triggers, the character image updates automatically (a random one from the set is set)

✨ new images for Dina, Jo and Fin

✨ each character got their own set of moans,

person: you can reassign them in the context settings

✨ Fin got his own branch in the prologue (first meeting in the doors event)

🛠 sandbox: fixed a bug with characters leaving; the leave option is left only for Dina, but she can always be brought back

🛠 story: fixed an error with the old man when trying to pretend you know him

🛠 fixed the translation of hints (3 quick replies)

🛠 apk: now when clicking on the main game screen with a character, a smooth view-shift animation is used

🛠 fixed the buttons and triggers for skipping sexualized scenes, added a sound effect**

🛠 sandbox: the back button now adds the previously sent text into the input field, works only 1 step back. fixed a background bug where changing a character's skin changed the existing location/generated background to the default one

🛠 gemini: now on hitting the limit or any other error, the previously entered text will be caringly added into the input field - for editing or resending

🛠 fixed a bug where opening the game set an incorrect endpoint for claude models

🛠 slightly changed the parallax

*my joy may seem strange for such a trifle (actually not a trifle :) - it's one of those things you keep putting off thinking it'll take several days, and then you wonder why it wasn't your first idea; turned out it's solved literally with one line,

the hero of this update is - white-space: pre-wrap

**false positives are fixed too. an important addition: when using the free cosmosrp version - due to the model's weaknesses, the triggers for when the user wants to leave or the character wants to end - may work incorrectly; in that case the button will definitely skip the dialogue

🛠 fixed an annoying critical bug with the protagonist's gender in the story, now it updates correctly depending on what's chosen on the main screen

🛠 increased the character size

🛠 fixed minor translation errors

✨ for Dina in the sandbox: slightly edited personality, updated images, now she can undress to a bra, can leave if offended or for any other reason*

✨ Fiona, Dina and Ava got images at the moment of climax

*if everything's fine with the leaving mechanic — it'll be added to other characters later

✨ reworked the prologue:

-- added the ability to play through the story part as a female protagonist. (Fin will be woven into the prologue's story in the next update);

-- replaced poor images;

-- in critical spots reworked the user-input windows so transitions are seamless;

-- polished some events;

-- added ambient sounds;

✨ apk: added the ability to copy text from the dialogue log (a temporary solution, downloading the file as on desktop will come later)

✨ a few new sandbox images for Jo

✨ added anthropic to proxyapi. (desktop)

✨ the game is fully translated into english

✨ added cosmosrp-pro (via a subscription to pawanosman's patreon)

✨ to apply a custom context you no longer need to tap the thumb

🛠 an attempt to fix empty messages when using gemini-1.5-pro models

🛠 fixed issues with narrator censorship

🛠 improved location-change detection in the sandboxes for english

🛠 now in all sexualized story segments, besides the special continue-story button, you can finish just by making it clear in the dialogue, and characters can also decide to end on their own

🛠 removed poor models and backgrounds from the start screen, added new ones

🛠 fixed a narrator bug in the english version

🛠 fixed a visual bug in the prologue's story part with the poems

🛠 fixed an interface bug with the help link from settings

🛠 fixed errors with proxyapi and proxy, on desktop fixed the claude+proxyapi error

🛠 reworked the poems event: fixed a critical bug that made it impossible to progress; nerfed from the required 27 points to 23. fixed the second-attempt output. now the event isn't mandatory and after 3 attempts the story continues, added a gallery image as a reward for completing it

🛠 due to performance issues the boring cursor was returned by default

🛠 fixed translation errors

✨ Jo learned to undress to a top + a few new images

✨ for base and person subscribers: added the proxyapi service, you can pay for openai and gemini with ruble cards, i'll add claude later. seems okay, BUT very expensive, feels like x3-x4, if paws or other options don't work and you have lots of money

🖤 linux build, AppImage - the analogue of the windows portable version

✨ Fiona: now can undress to a top; can take a ride in a car; changes in personality

✨ in the prologue of the story part, in the nighttime event with Fiona, added the option to chase her off + she can now leave on her own if she wants

✨ redesigned pages: help, settings, save/load - now with transparency and blur they cover the main game screen

✨ now when the input field appears, the background blurs a bit

✨ apk: now instead of a black background - a blurred something repeating in the colors of the currently used background

✨ added sounds to the interface

✨ added the chatgpt-4o-latest model, deprecated gpt-4-0125-preview and gpt-3.5-turbo-0125

✨ gemini: you can now enter several keys, comma-separated; when the limit is used up the next key will be used. an in-game proxy will come later; changed the system parameter topK, toward a larger word selection at each step

🛠 the character image/animation became a bit bigger

🛠 set up statistics from umami, instead of yandex.metrica*

* GDPR-compliant, doesn't violate privacy, collects no personally identifiable information and anonymizes all collected data

✨ a mechanism to erase part of the system context for all sandbox characters

✨ added the gemini-1.5-pro-exp-0827 model

✨ a few images for Ellie and Fiona, clothed

🛠 fixed the system contexts of all sandbox characters

🛠 optimization: downscaled heavy animations, images

🛠 a small redesign of the main screen

in this part there's no sex, more narrative and lore, to push the story further

✨ story part

✨ a new character (unlocks after completion)

✨ person: the ability to erase memory and context for the new character; a small story navigation

🛠 apk: fixed video display

🛠 comics mode: added text hiding on "h"

✨ characters in the sandbox can now dress

✨ sandbox: added a choice from a set of clothed character skins

✨ added the llama-3.1-8b-instant and llama-3.1-70b-versatile models

✨ added streaming output for gemini*

✨ added a streaming-output setting**

✨ added a font-size setting. by default it adapts to the amount of text, when off the font size is fixed

✨ person: added a setting to generate a short summary from message history, every n messages

🛠 max freedom-level value increased to 0.99

🛠 tweaked the contexts

🛠 fixed a small reward for winning the "Save the princess" minigame in the prologue of the story part

🗝️ changed the boosty/patreon password

* for now only with your own vpn, an in-game proxy will come later

** for all except anthropic models

✨ added the experimental gemini-1.5-pro-exp-0801 model

✨ added the club location

✨ new images for Jo

✨ in the prologue expanded the doors event: after dinner, added the option to drop in on Fiona

✨ added the small claude-3-haiku model

✨ rewrote and optimized the contexts of all characters

(on weak models it should fix the context-overflow error. on big ones - individual traits are a bit more pronounced)

🛠 fixed a bug when using gemini/claude in the story moment after the walk with Jo

🛠 temporarily deprecated the llama3-groq-70b-8192-tool-use-preview model

✨ added the park location

✨ gemini: added two backup models 1.5-pro and 1.5-pro-001, in case google's outage repeats

✨ added models: llama3-groq-70b-8192-tool-use-preview, gemma2-9b-it, gpt-4o-mini

✨ background generation from a text description (button 1)* now works on dalle-3, the images became widescreen, the side margins were removed, the quality grew significantly and became a bit faster

✨ person: comics mode in sexualized scenes

🛠 fixed a bug where cosmosrp was set as the default model on game restart (try 2, now for sure)

🛠 gemini: for gemini-pro models, increased the delay before retrying to get an answer on error, now the message-limit notice should appear less often

🛠 now after undressing a random image/animation from the character's set is shown, and on a repeat activation it isn't reset

🛠 font size increased and adjusts to the amount of text

🛠 for image generation from a text description the correct openai endpoint is set, now it doesn't depend on the model selected in settings and can be used simultaneously with other models

🛠 lowered the bitrate on a few tracks, for optimization

🛠 redrew poor models from the story

🛠 optimized the sandbox characters' contexts

🛠 refactoring, a small ui redesign and minor bug fixes

✨ locations were introduced into the game (in the sandbox)*: shower, restaurant/cafe, forest, field, beach, library, desert, subway, moon, mall, city, hospital, apartment

✨ added a list of available locations to the help section

✨ a small update for Laura in the sandbox

✨ changed the look of the user window toward minimalism

✨ sandbox: all characters became a bit less horny at the start of a dialogue; turned off avatars next to the character's name

✨ person: added a clear-message-history button, the saved history became a bit more informative

🛠 redrew poor images and models from the story

🛠 fixed the cosmosrp issue when other fields aren't filled, now on first opening the game cosmosrp works correctly out of the box, without extra setup

✨ added a neural net from pai - cosmosrp, one of a kind, requires no api key

✨ added neural nets from anthropic

✨ update for Dina: images, animations and a small system-context adjustment

✨ gallery: in the story reworked the triggers for opening gallery items, added new items.

✨ a small update for Jo

✨ added backgrounds

🛠 the block bypass now works correctly when picking gpt models from settings, previously you had to press bypass again

🛠 now in settings you don't need to press the thumb, keys and password are saved on filling. (for contexts the thumb stayed)

🛠 slightly changed Dina's context in the sandbox

🛠 the long-suffering block-bypass setting is finally fixed and named more concisely "proxy". works with gemini, claude and chatgpt

🛠 optimized the animated backgrounds

🛠 fixed the cursor and the load-save screens

🛠 minor interface tweaks

🛠 apk: links from the game now open correctly in the browser

🛠 apk: the exit button disappeared from the main screen as unnecessary

✨ person: added the ability to unlock all gallery items in settings

✨ added llama, mixtral and gemma models via the groq api

✨ added a setting so that on launching the sandbox the character starts the dialogue first (a small situation is generated)

✨ added the first male character, Fin, to the sandbox

✨ added a button to the log to save the message history to a text file

✨ added model tooltips

🛠 fixed a bug with a random Dina image on sandbox start

🛠 fixed a bug: exiting to the main menu kicked you to the start of the game

🛠 apk: fixed a critical bug on some phones where only Laura was available from the main screen. now instead of a carousel a static block with all characters

🛠 fixed a bug with incorrect display of the dnd scenario setting state

🛠 brought the in-game help section up to date

🛠 option generation now works for Ellie too

🛠 added backgrounds, removed poor ones

🛠 a small interface redesign/refactoring

✨ a new sandbox character: a grown-up Ellie from the last of us, by a subscriber's request (images only for now, no animations).

✨ added hints (3 quick replies) for gemini, now it works the same as with gpt

✨ person: for gemini added history saving (every 11 messages a short recap of what's happening is formed)

🛠 improved critical-error output, it no longer causes a hang. overall it became more stable (especially with gemini) and friendlier to new users

🛠 fixed issues with incorrect writing and saving of the block-bypass setting

🛠 the setting with hints enabled is now saved and doesn't reset on opening the game

🛠 fixed the mini heart animation on undressing

🛠 gemini: fixed a bug in the story where characters got confused and lost the context of what's happening, now it works the same as with chatgpt and history is passed correctly

🛠 gemini: fixed a bug where messages in the story weren't shown and were only visible in the log

🛠 fixed a bug where for image generation you had to select the chatgpt model, now the key is taken correctly from the openai field

🛠 removed poor videos from the intro, replaced with images

🛠 the moan volume now obeys the voice-volume slider in settings (sorry if anyone got into an awkward situation because of this)

🛠 fixed a bug in the prologue story at the table

🛠 fixed narrator-text display when using gemini (a Fiona image was shown by mistake)

✨ a small redesign of the main screen

✨ apk: character sandboxes are now in a carousel

✨ added the gemini-1.5-flash-latest model

✨ redrew poor images/animations for Laura, Fiona and Jo

✨ new images for the main screen

🛠 rewrote and optimized the characters' system contexts

🛠 reworked error output, became a bit "friendlier" to the user

✨ person: added a new model gpt4o (couldn't fully bypass censorship yet)

✨ person: sandbox: added user context to settings (you can write any info about yourself, personality, appearance, hobbies, how you see yourself or would like to be seen)

✨ added gemini

✨ desktop: added speech synthesis from openai (uses the openai api key)

✨ desktop: added speech synthesis from google

✨ characters now format their thoughts as (💭 ..)

✨ updated the image-animations for Laura in the sandbox

✨ reworked the model-selection block in settings and updated the chatgpt model list

✨ changed the yandex speech voice to marina and slightly increased the speed

✨ pai (added because it's "free and maybe later it'll be okay". at this stage almost unusable, poor quality, hallucinates and often spits out nonsense)

🛠 the personal chatgpt key works with the block bypass

✨ story part

✨ voice recognition, works via the user's browser, desktop only for now

✨ speech synthesis yandex speechkit api, requires a key

✨ the in-game help section reworked and added to the bottom game menu

✨ background generation, background and character variations. but don't get your hopes up, for now it's pretty trippy and creepy :)

✨ random clothing in the main menu

✨ new images and animations

✨ all the music in the game became looped, sorry i didn't do this earlier )

🛠 on mobile the font sizes and margins became smaller

🛠 fixed a bug with opening links on the main screen, now it opens in the user's browser, not in the shell

🛠 fixed interface bugs

🛠 updated the boosty password

✨ the "Save the princess" minigame can now be played in the sandbox with Dina

✨ added images to the gallery

✨ added sounds

✨ new animations

✨ lesbian contexts for the girls in the sandbox, activated by choosing the gender on the main screen

✨ added random images to the sandboxes, to bring in a bit of variety

✨ added the gpt-4-0125-preview (gpt-4-turbo) model, should be many times cheaper than gpt-4, but heavily censored for now

🛠 fixed the backend part of key issuing, became more stable, without gaps

🛠 fixed a bug where the log was shown empty

🛠 redrew poor images (intro part, sandbox)

🛠 nerfed the poem points from 30 to 27

🛠 hidden parameter top_p changed from 0.91 to 0.9

✨ the game translated to english (for now by a crude method using gpt4)

✨ hidden parameter top_p, as an experiment, changed from 0.89 to 0.91 (controls answer diversity by limiting the next-word choice to a subset of the most probable words. the lower the value the more predictable and consistent the answers, works together with temperature, which is in settings)

✨ rewrote the system contexts of the characters in sandbox mode. the characters' personalities became a bit more pronounced. a bit, because gpt3.5 is still weak at russian and the context length is limited, for now it's like this, it'll improve gradually

✨ added a diminutive or any other form of the name to settings (used in the sandbox)

✨ person: memory in sandbox mode, the number of messages stored in memory increased from 9 to 12, + every 11 messages a context is formed storing a short recap of what happened.

on opening the sandbox, if there's saved history a hint button "Where did we leave off?" appears, to remind you what happened

✨ story part

✨ added windowed mode, enabled in settings

✨ improved animated character images, with Jo as the example. the rest of the images will be converted in a similar way

✨ user context windows for characters doubled, from 1800 to 3600 characters

✨ a sandbox for Jo

✨ added the ability to choose a cursor in settings

✨ added a couple of animated images for the Fiona sandbox

🛠 fixed story bugs, including the critical "morning in the hospital with Fiona"

🛠 optimized animated images and videos

🛠 added music in places where it should have been; on some tracks the bitrate was lowered for optimization

🛠 optimized the output of extra images on undressing, on weak machines it caused lag

🛠 did a planned refactoring

🛠 reworked a small story bit after the "explosion"

🛠 changed the boosty password

- fixed a bug with line trimming

- fixed an annoying bug that broke the narration and led to a chunk of raw story

- enlarged character images on mobile in portrait mode

- fixed a bug with hint overlap in the interface

- added a save button 👌🏻 for custom contexts; previously they were applied on re-entering the sandbox or calling images on undressing

- added a sandbox mode for Fiona

- added an openai block bypass so it works without a vpn from Russia and Belarus - enable in settings

- fixed a critical minigame bug that prevented it from launching, reworked the output to iframe, which gives much bigger possibilities for interacting with the main game engine (when the time comes)

- the ticker now updates online

- replaced Fiona's models from the intro, the rest will be updated later

- added a sandbox mode for Dina

- fixed the image-activation mechanism for Dina

- fixed a bug with extra images when there are too many and they don't fit the screen

- nerfed the poems a bit, now you need 30 points instead of 31. the analysis became shorter

- fixed some story transitions and interface bugs

- image picking now activates when a character is undressed, added images

- changed the default model to gpt-3.5-turbo-0613

- apk: added portrait mode for mobile, still rough, but typing is more convenient than in landscape, it'll be refined

- added the ability to choose from 3 models in the public build

- added a public sandbox mode for Laura, jump from the main screen by clicking Laura, message history stored in memory is limited to nine

- added a visible scrollbar in the message history

- changed the look of the user-input window

- fixed interface and mobile-version bugs, did a refactoring

- added a ticker with important info. later it'll become online, important and urgent info will be shown in this ticker

- for person subscribers: added the ability to write your own context

- story part

- added backup addresses with keys and error handling, now in case of a block, a spoiled key doesn't get in the way of seamless progression

- to test a hypothesis the server side moved to a different domain, users from Ukraine pls check whether it works without a vpn

- updated the interface, start screen, custom cursor

- apk: fixed dialogue overlap on user input and hints

- added a button to skip long dialogue sessions, activated at the start of a dialogue at the moment of showing user input

- introduced a hidden parameter top_p at 0.81 (affects word selection and answer consistency, 1 - no filtering applied, 0 - selection strongly limited), in search of the ideal value, it'll change

- reworked the poem-writing event, lowered the passing score, added an alternative way to reach the event (better to load shortly before the end of the last dialogue with Laura, before the "explosion")

- fixed minor bugs and contexts

- added analytics, counts only the number of users, collects nothing else

- added music

- chatbot text output reworked to streaming, response time dropped to a minimum, in most cases the answer is 0.1-0.5s. Yahoo!

- the splitting of long dialogues is now shown following the streaming output, bug or feature - more of a bug, but in my view the user experience doesn't suffer and there are certain tech difficulties with dropping this solution for now

- for subscribers, added the ability to change the model in the game settings

- apk: the user input is now always above the dialogue, solving the big-virtual-keyboard problem

- minor interface changes, added smoothness where it was lacking

- added parallax

- added a model freedom-level (temperature) setting, from 0.1 to 0.9 with a detailed description

- as an experiment changed the working model from gpt-3.5-turbo-0613 to gpt-3.5-turbo

- in the event with Jo (wifi) redrew poor sprites and added images. the rest will be redrawn periodically too

- slightly increased the number of dialogues stored in chatbot memory besides the system contexts

- adjusted the splitting of big dialogues, it now happens less often

- tweaked the characters' contexts

- fixed a bug where the game had to be restarted

- fixed errors during the appearance of narrator dialogues and when characters start a conversation without user input

- changed the save-loading logic, now old saves won't throw errors and most of them will keep working in cases where old content changed

- long character dialogues are now split into several smaller ones, which works better for immersion and looks more natural than walls of text

- added the start of the main story branch (getting there is fairly easy, so on release there's no hint at all, so as not to spoil the "pioneer" impression)

- fixed a text-trimming error

- redrew poor images of Fiona

- videos re-encoded to webm for optimization and to solve an availability problem on some android versions

- fixed a context-duplication error when using the "back" button, which caused characters to stop answering relevantly to the latest messages and broke some constructs

- the characters' chatbots learned to talk to each other (test-only for now in one small dinner scene)

- did a refactoring, the build became much more stable

- added a log (in the simplest form, will be refined)

- added events with Laura, on the morning of the second day

- fixed interface errors

- reworked and optimized the characters' contexts

- added the ability to enable/disable hints (3 quick replies) in settings, which affect loading speed

- fixed the token issue for the chatbots, now everything is more stable and hopefully for a long time 🤞

- handled critical errors that led to crashes

- fixed saves

- fixed crashes due to message overflow

- added a small intro

- reworked and optimized the characters' contexts, the neighbors can now take offense, the characters' speech became more natural and varied