Where do games disclose AI use?

This game's cat characters are based on art assets we created, expanded with generative AI into a variety of patterns and colors. Generative AI image tools were also used at the concept stage for the store page's capsule images and icons. Players can also optionally register their own free Google API key to chat with their cats or check today's fortune. This feature connects live to Google's Gemini API while the game is running to generate responses; the rest of the game can be enjoyed the same way without registering a key.

Companion OS uses local LLM models and supported cloud AI providers to power some companion conversations and interactive responses.

Meeting the World (Aira) uses generative AI to create conversations and optional synthesized speech during gameplay. Depending on the enabled features, AI can also select predefined character actions and influence conversational emotion, autonomous behaviour and long-term understanding. The Windows build includes a local Gemma 4 e2b chat model. Players may optionally configure their own external AI provider. Local speech synthesis uses CosyVoice3 with a bundled reference voice or a voice reference selected by the player. AI responses may be inaccurate or inappropriate and should not be treated as professional advice.

Used in simple artwork, NPC chat systems, tactical realism systems, enhanced system design, economy management, debugging and design assistance.

Using AI to offer players more possibilities in chat

Part of the game's artwork - backgrounds, character illustrations, and CG images - was created with the assistance of generative AI tools, and some background music tracks and character voice lines were produced using AI music and AI speech generation services. All AI-assisted assets were reviewed and selected by the developer. The story, characters, and writing were authored by the developer. The game also includes an optional runtime AI dialogue feature that runs on a local model bundled with the game, fully offline; it generates short character replies only and cannot change the story or endings.

Generative AI tools were used to help create some of our promotional artwork, such as the store capsule and key art. AI is used to develop intelligence of the enemy bot players in the arena. It was reviewed and selected by our team. No AI content is generated during gameplay.

During matches, AI characters generate text through our official service, which connects to a third-party AI provider. Players do not need to enter an API key or manage a third-party account. The game includes a limited number of Turing Tickets; creating a match normally uses one. Additional Tickets are planned for purchase through Steam in-game purchases, and some may be earned through gameplay. Generative AI was also used to create promotional illustrations and store artwork. These staged images are not gameplay screenshots.

This utility app uses artificial intelligence models to perform real-time speech-to-text transcription (ASR), live subtitle translation (MT), and local text-to-speech synthesis (TTS) based on the user's voice or PC audio stream.

This game uses a locally running AI model to analyze music added by the player and generate playable rhythm-game note charts. The AI generates only note timing, lanes, types, and durations; it does not generate music, images, voices, text, or code. Music added by the player is not sent to any external AI service or developer-operated server.

1. This game uses AI to interact with the input text commands. 2. Some of the art resources and music assets in the game are generated by AI during the development stage.

Deepwood Valley uses a generative AI language model to power conversations with the characters who live in the valley. The model runs entirely on your own computer — no account is required, and nothing you type is sent to us or to any third party. Every line of AI-generated dialogue passes through a deterministic content filter before it is displayed. This filter is compiled into the game, runs on all generated speech without exception, and cannot be turned off by the player or by mods. Character personalities, story, world design, and all art and audio are authored by our team.

This game utilizes generative AI technologies in the following areas: [Pre-generated Content] Part of BGM: AI-generated audio is temporarily used as placeholder assets during development (to be progressively replaced in future updates). [Real-time Generation during Gameplay] In-Game Text: Dynamic generation of dialogues and narrative text tailored to in-game situations. Character Voice: Real-time speech synthesis for character voices based on generated text. *All real-time generation processes run entirely locally and securely on the player's machine, with no data transmitted to external servers. Furthermore, there is no feature allowing players to freely input custom text or prompts.

Generative AI was used during development for some visual and audio assets. During gameplay, narrative text and character dialogue are generated in real time by a third-party AI model selected and configured by the player.

This game uses generative AI for some pre-produced visual assets, including character artwork and backgrounds, and for text generated during gameplay. A language model running locally on your computer generates VTuber dialogue and responses based on player input, character settings, and the current game situation. Generated dialogue can vary between playthroughs.

Generative AI was used to assist with the development of this game and with the creation of some of its visual assets. The game includes an optional custom character feature. Players may enter text descriptions and provide reference images to generate character visuals. Players may also use text descriptions together with character reference images to create custom cinematic videos for their characters’ special moves. These materials are generated by generative AI through an online service during gameplay. Once generation is complete, the resulting content is provided to the player for use as part of their custom character data.

Desktop Companion uses generative AI: your companion's replies are generated live by a large language model that runs entirely on your own PC. The app ships with its own safety-tuned model; nothing is sent to an external AI service. Built-in, application-level safeguards screen generated content to block prohibited output.

Shiny Gen ships over thirty original games and scenes. Their art, animation frames, music and sound effects were produced with the same AI generation tools the product offers to players, using the third party services listed in our Live-Generated answer. All of it was directed, reviewed and assembled by us before it shipped, and the game code itself is authored and hand edited by our team. No pre generated content in the shipping build is produced from a player's input, and none of it is generated at runtime from our side. Shiny Gen is a game creation tool, so live generation is the product rather than a feature of it. When a player asks for something, we call third party AI services to generate game code, images, animation frames, music, sound effects and 3D models from the text they type. Generation always happens at the player's explicit request, never in the background, and always requires a free Shiny Gen account and an internet connection. Every request goes to an established commercial AI provider: Anthropic, OpenAI, Google, OpenRouter, Cloudflare Workers AI, Runware, Retro Diffusion, Replicate and ElevenLabs. Each request is governed by that provider's usage policy and runs under the provider's own built-in safety systems at their default settings. We never disable, weaken or opt out of a provider's safety system anywhere in our product. OpenAI image generation runs under OpenAI's default moderation, and Google image generation carries Google's SynthID provenance watermarking. Guardrails on our side: - Our Community Rules, which are binding under our Terms of Service and published on our website, set out what may not be generated or shared. They prohibit, among other things, any sexual content involving a minor and sexual imagery of a real person without their consent, and they treat a prompt asking for prohibited content as a breach whether or not the model produces it. - Generation requires a signed in Shiny Gen account. There is no anonymous generation, so every request is attributable to an account. - Generation is rate limited per account and per address, and it is metered against a paid credit balance, so volume abuse is bounded rather than free. - The prompt and the resulting asset are retained against the account that made them, which is what lets us act on a report. - Anyone can report content to support@shinygen.ai. A person on our team reviews each report and can remove the content and close the account responsible. - The in app feed cannot receive player made content. It is served from our own catalogue and is enforced in code: the publish endpoint refuses any caller that is not our own account. Nothing a player generates can appear in another player's feed. - Multiplayer contains no free text channel of any kind. Communication is a closed set of eight fixed emotes, so there is no user to user text surface to moderate. - Player made games are never separate programs. They run as interpreted scripts inside our own sandbox, which is an allowlist: it has no filesystem, network, process or operating system access, and no reflection or dynamic dispatch. A generated game cannot execute native code, start a process, reach the network or read the player's files. We do not permit and do not support the generation of Adult Only sexual content, and we will act on reports received through Steam's in game reporting system and at support@shinygen.ai.

Lost Gaijin uses generative AI during development and gameplay. Selected pre-generated visual and audio assets used in the game and its Steam materials were created or edited with generative AI tools and reviewed by the developer. During play, text-generating AI responds to player input and game state to create dynamic story content and personalized Japanese-language coaching. Live-generated text is provided through OpenRouter and may vary between playthroughs.

Q-Rely uses locally running artificial intelligence to provide interactive AI-generated responses during use. The AI can generate text responses based on the user's input and, depending on the user's configuration, can also generate voice output using locally installed Text-to-Speech (TTS) engines. Generative AI was used during development to assist in creating the application's logo. The logo is a fixed visual asset and is not generated or modified by generative AI during use of the application. Q-Rely does not use generative AI to create the application's default character artwork, backgrounds, or other visual assets during use. The application's AI-generated content is primarily conversational text and, when voice output is enabled, synthesized speech. Q-Rely is designed to allow users to run AI models locally on their own computer. Users can select and install compatible AI models and configure the application's AI and voice settings according to their preferences.

Xuanji · Diaochan uses a live generative AI to produce Diaochan's dialogue and spoken responses in real time. Every line of conversation and every spoken reply is generated by the model during play, not pre-written. The character artwork, scene illustrations, music, and sound effects in this game are created by our team without generative AI. Guardrails in place: A historical-fiction persona that keeps Diaochan in character as a 2nd-century figure and prevents her from speaking as a modern AI assistant. Retrieval-augmented responses grounded in a curated corpus of Han dynasty historical texts, so factual claims can be checked against primary sources. Pre-filtering of user inputs and post-filtering of model outputs against blocklists for sexual, violent, and politically sensitive content. An in-product report button on every response, plus mute and memory-reset controls. Players can also report any in-game AI response they consider inappropriate through the Steam overlay.

We build Last Treasure Online with AI tools. Much of the code is written with AI, including shaders, scripts that model props and furniture in Blender, and scripts that synthesize sound effects and ambience. Most text in the game is AI generated, but all names, personalities, schedules, stories, lore, knowledge, etc is strongly scaffolded and enforced by us as designers. In play, NPCs speak through an AI language model on our server in order to talk about hundreds of subjects in different ways depending on what they know about them, you, the world, etc.... The model only rewords lines and facts we wrote: it cannot invent places, items, prices or quests, and players never type text into it. No images, textures, 3D models, music or sounds come from diffusion image, 3D or audio models. Characters and clothing are hand-painted, and the world's textures are photographs.

Some in-game models and music were created with generative AI. Our original cover illustrations and title lettering are handmade; selected store-art adaptations and the app icon used AI tools. Optional chat uses a lightweight AI model running entirely on your computer to answer small everyday questions. You can turn chat off at any time.

Generative AI assisted with SystemChart's branding and store descriptions. The optional assistant uses a user-configured model to select diagnostic tools. In the Steam app, SystemChart writes the displayed summaries from validated tool results; free-form model text is not displayed. External AI clients connected through the optional tool server control their own responses. Core capture and charts do not require AI. No model or hosted inference service is included.

Generative AI assisted with selected original visual assets used by Tater Tube Player, including mascot illustrations, fictional demo artwork, Discovery category art, and Tube TV channel logos. These assets were reviewed before inclusion and do not contain third-party movie or television artwork. During use, the optional Tater Picks feature can generate short recommendation explanations and a synthesized voice briefing based on titles in the user’s own Tater Tube Server library and viewing history. Tater Picks requires the user to configure Tater Link separately, does not accept free-form prompts through the Player, and is hidden when not configured. It does not generate movies, television programs, music, or other playable media.

RealSub uses AI technology as its core function: it runs speech-recognition and voice-activity-detection models (OpenAI Whisper via faster-whisper, and Silero VAD) locally on the user's own computer to turn audio the user is already playing into live subtitles. If the user enables translation, the recognized subtitle text is sent to the translation service the user selected (Microsoft Translator by default, or an OpenAI-compatible LLM endpoint, DeepL, or a translation server the user runs locally) and the returned translation is displayed. Audio is never uploaded. The subtitles and translations produced this way are generated at runtime, from the user's own audio, and are shown only to that user on their own screen. RealSub does not generate art, music, voice-over, or narrative content, and it does not share generated text with other users or publish it anywhere. None of the shipped content assets are generated by AI: interface text, icons, store artwork, screenshots and video are written, laid out or recorded by us. The store capsules are programmatic layouts, and the demo footage is real screen recording of licensed material.

This game leverages AI content for the computer player.

Vectorline uses on-device AI for its optional voice features. When you speak commands to aircraft, a local speech-recognition model transcribes your voice into ATC instructions. Pilot and controller radio readbacks are spoken aloud using a local neural text-to-speech model. Both models run entirely on your own machine — no audio, text, or data is ever sent to an external server, and no internet connection is required for these features. The spoken content is limited to standard air-traffic-control phraseology generated by the simulator itself. You can disable these features in the settings without impacting gameplay.

This game uses a large language model to generate the AI companion’s dialogue and ending titles in real time. Dialogue responses are generated based on player input and conversation context. Generative AI tools were also used to help create or edit some store and promotional artwork and assist with text translation and localization.

The game uses generative AI to create molecular structures during gameplay. Molecules are procedurally generated by a generative AI model embedded locally within the game and operating as part of the game’s molecular design system. The in-game generative AI system is used exclusively for molecular structure generation and does not generate dialogue, artwork, music, or other narrative content during gameplay. During development, generative AI tools were used to assist with artwork preparation, text polishing, and UI code development.

This application essentially installs and runs free, open-source artificial intelligence models on your computer. Its working principle is as follows: the model listens to the video you are watching 5-10 seconds beforehand, then generates real-time dubbed audio in your chosen language, and simultaneously produces subtitles in your selected language.

Tug-O-War Wrestling uses a locally-run AI model (Llama) to control enemy wrestler behaviour and decision-making during matches. This AI is not used to generate any art, audio, dialogue, or other player-facing content - all in-game content is human-created.

All in-game visual and audio assets are procedurally generated by code at runtime. The build contains no pre-rendered AI, painted, or recorded media files. AI (Anthropic's Claude) was used during development only to help draft source code and in-game text, which was manually reviewed and edited. No AI services are accessed while running the game. The only third-party assets are two open-license fonts designed by human artists.

The game will connect to an open-source local large language model that is already built into the game upon first launch, and use it during gameplay to generate NPC dialogues, host responses, and reasoning review feedback. Players may also choose to access compatible third‑party large language model APIs. The text generated at runtime is based on player questions and current story data.

Starflare Dominion uses generative AI in two ways. Some visual assets, music, and sound effects were created with assistance from generative AI tools and were selected, edited, processed, and integrated by Whimstead LLC. During Story play, hosted OpenAI models may generate bounded narrative text, including pilot backgrounds, quotes, Cantina news, Old Man Rich stories, campaign consequences, and final legends. Players do not directly prompt an unrestricted chatbot. Generated text is checked by structured rules and validators, with retries and curated local fallback content when needed.

This app includes an optional AI assistant that helps interpret game server console output and error logs. When you send it a log fragment or a question, the text goes to our backend, which forwards it to Anthropic's Claude API and returns a generated explanation with suggested fixes. The assistant runs only when you invoke it — nothing is sent automatically, and the feature is off by default. Responses are generated by a language model and may be inaccurate, so verify any suggestion before applying it to a live server. No other content in this app is AI-generated.

English: This game uses generative AI to create case stories, suspect dialogues, and investigative clues in real time during play. AI is provided by third-party services — you bring your own API key (entered in the in-game settings). No microtransactions.

AI assistance was used in the development of this game. Players are strongly advised to play using mainstream free "flash" AI APIs such as Agnes API or Qwen API — simply fill in your API information in the game settings. For the sake of speed, enabling thinking mode is not recommended. This game contains optional generative AI features. In “AI-Enhanced Narrative Mode”, the game calls the third-party LLM API you provide yourself, generating more immersive event text in real time as you play. This mode is entirely optional: even without connecting any API, you can still fully experience all of the built-in, human-written historical content and the core gameplay. The game will not use your gameplay data to train any models. All AI-generated content is controlled by your own API key and quota, and you are responsible for your own usage.

This software incorporates a local AI model, which is used solely to provide an "intelligent semantic search" service. When you enter a text description, the AI analyzes the semantics in real time and searches your local image library to display the most relevant image results. All search results are sourced from your local image library; the AI does not generate or create any new image or video content. Users can choose whether to enable this feature.

Pre-generated: Some 2D illustrations and the pre-recorded narration were produced with AI tools and reviewed by the developers before inclusion. Live-generated: An on-device language model (no internet connection) writes in-game text such as narration, dialogue, and the names and lore of items, relics and skills, and selects among game-defined options for certain run content. Damage, stats, difficulty and progression are governed by the game's own rules. Generated text is checked before it is shown, and no art or audio is generated at runtime.

The bartender's side of every conversation is generated live, while you play, by a language model that ships inside the game and runs on your own computer. Nothing you type or say is sent anywhere. There is no account, no server, and no internet connection required — the game works with the network unplugged. When the character speaks her lines aloud, that audio is synthesised locally by the same install. Everything else in the game is authored by hand: the world and its writing, the cast of regulars, the drink recipes, the art, and the rules the bartender plays by. Because her dialogue is generated rather than scripted, it cannot be listed in advance. The controls on it are: a fixed authored prompt that defines the character, the setting, and what she will and will not do; a filtered generation pipeline; and a specific model chosen and regression-tested against an automated suite of behavioural checks before it ships. Anything the character has remembered about you can be viewed and permanently deleted from inside the game at any time.

Quiet Sky uses generative AI during gameplay to simulate evolving societies and create narrative observation reports, summaries, Companion responses, and optional report illustrations based on the current simulation and the player’s decisions. The Steam desktop edition runs the included generative models locally on the player’s computer. Generated results vary between campaigns.

This game uses generative AI in a limited capacity on the game server in real time. The AI is used solely to generate narrative text: mission names and descriptions, in-world news broadcasts, and some NPC dialogue. All game logic — including mission mechanics, rewards, and balance — is governed by deterministic code and is never decided by AI. If the AI service is unavailable, the game automatically falls back to pre-written text templates. Portions of imagery were produced with the assistance of generative AI.

Lumi uses generative AI to create dynamic conversations and spoken responses in real time. AI inference, speech recognition and speech synthesis run locally on the player's PC. AI model files are downloaded during setup, but player conversations and generated responses are not sent to an external AI service. Available capabilities may vary depending on the locally selected model.

We use AI to generate interactive videos in the game.

Text-based conversations with 12dora in-game are produced in real time while the game is running, powered by an integrated large language model (LLM).

Asiden uses generative AI to create companion conversations, custom personality text and spoken responses during use. When screen awareness is enabled, AI can interpret screen content to inform its responses. Local AI features run on your computer. Optional cloud features send the information needed for the selected feature through Asiden’s managed service to external AI providers. Generated responses may be inaccurate and can include mature language, humour and non-explicit flirtation.

Take a Joke uses generative AI in a few clearly defined places. Created during development and shipped as fixed files: • Artwork — the game's logo, icon, and character avatars were created using AI image-generation tools, then selected, edited, and assembled by hand. • Narrator voice — the spoken introduction is a synthesized voice performance created with ElevenLabs, reading a script we wrote. • Cards — the great majority of the prompt and answer deck was written by hand. A small number of cards were drafted with AI assistance. Every card in the game was read and approved by a person before it shipped. Generated during play (optional features): • The narrator occasionally reacts to what is happening in a round. The lines come from a fixed, human-written library; a player's name is inserted and the line is voiced with ElevenLabs. • The optional end-of-game awards ceremony writes a short, custom awards show about the game you just played — the award titles and descriptions are AI-written from that game's results, then voiced. When those optional features are on, the game sends player nicknames and submitted answers to the services that generate them. With them off, nothing is generated while you play.

This game was created by a solo developer. During development, generative AI was used to assist with some artwork, store promotional images, localized text, and narrative materials. While the game is running, it connects to external AI services to dynamically generate Froggy’s dialogue, travel letters, postcard images, and parts of its travel stories based on player conversations, character memories, real-world weather, and current travel status. Generated content may vary depending on player input and service availability.

Dream Simulators uses generative AI to create dynamic dialogue, story continuations, characters, missions, and narrative events based on player input and the active Lore system. Each Expansion Pack features its own world, Lore, characters, and adventures, creating a unique storytelling experience. The AI also generates creative scene descriptions and prompts that players can use as inspiration for creating images and videos. Because the experience is driven by player input, generated content may vary between playthroughs.