One AI, four stages: role, photo, debate, crisis
speak German with AI
At night, the group is often empty or moves too fast. AI isn’t a permanent replacement for people. It’s the moment that happens anyway. Four modes, one microphone.
Role
Bakery, government office, train station. I play one side, the AI the other. Short sentences. “Please,” “Repeat,” “the price,” “the form.” I stay in a scene until the line is right, and I don’t change my “costume” every thirty seconds.
Photo
A picture from my own gallery, kitchen, or street. The AI asks for the location and subject. This is closer to a B1-level description than a stock photo from a textbook. One camera permission, one picture per session. Not the entire role.
Debate
Here, someone wants a reason—“because” and “even though.” I choose a topic on which I have an opinion—rent, a map, or a bar—not a topic I only know because it’s in a chapter. The AI is allowed to disagree. Otherwise, it’s just a monologue with a microphone.
Situation
Train gone, burst pipe, check-in. Practice at home so the counter isn’t your first attempt. The same short sentences as in the role-play, only with more tension in the content and less in the room.
What the AI Isn’t
It doesn’t sell token packages in this app. You can use your own Gemini or OpenAI key, which stays on the device. If you don’t have one, you can still practice. The four modes are the path, not a store.
People come next: language group, list, meeting. The AI warms things up. It doesn’t replace the voice of the person across from you.
How I Spread Out the Four Modes Over a Week
Monday: Role-play, one scene, ten minutes. Tuesday: the same photo as last week, two different sentences, so I don’t have to look for a new picture every time. Wednesday: a short debate, a “because,” an “although,” then wrap it up before it turns into an essay. Thursday: a situation I really dread—at work or on the train—just the first four sentences. Friday: people, if my mouth cooperates. If not, back to the role, the same scene, not a new one.
I don’t switch modes in the middle of a sentence. The transition is the moment when I stop speaking and start organizing my thoughts. One mode per session. The key remains. In the evening, the words from the scene are transferred to their own card, along with the example sentence I spoke—not a sentence from someone else’s list.
Anyone who “runs through” all four modes in one evening ends up with four beginnings and no ending. The app allows that. I don’t. The group on Friday gets one sentence from the week, not a summary. If no one is in the group, the sentence stays with the AI. It has still been said. That counts for more than a plan waiting for a full round.