She Doesn't Speak My Language

Character art, event illustrations, backgrounds and gift pictures were generated with open-weight image models (Animagine XL 4.0, Qwen-Image-2512 and Qwen-Image-Edit-2511) under the developer's direction, from a fixed design sheet for each heroine, and checked for consistency (same face, hands and outfit) with automated tests and visual reviews. The heroines' voices are synthetic: designed from written descriptions with open-weight speech models (Qwen3-TTS, Maya1, FireRedTTS3), not cloned from any real person. Background music was generated with ACE-Step 1.5; some sound effects with Stable Audio Open 1.0. The story, characters and dialogue were written with the help of AI writing tools under the developer's direction. During conversations, a small language model that runs entirely on your own computer (Qwen3-4B-Instruct) phrases the heroine's replies inside story scenes designed in advance; it does not decide the plot. It also writes translations when you ask and short coaching notes after a date. Unrecorded lines are voiced on your computer (Kokoro-82M), and speech you say into the microphone is transcribed on your computer. Nothing you type or say is sent over the internet. Every generated reply is checked while it appears and when it is complete; replies with sexual wording, hateful language, mixed languages, references to being an AI or to its instructions, or that ignore what you said are removed, and if no reply passes, the heroine says a pre-written line.