Capture first, organize later
Dreams disappear quickly, and typing a polished entry is often too much friction first thing in the morning. Tranquili-Tea starts with voice: the user describes whatever they remember, even if the story is fragmented or strange.
Whisper transcribes the recording, GPT-4o Mini creates a concise summary, and the resulting description becomes the basis for a generated image. The visual is not meant to be literal; it acts as an evocative cover for the memory.
Record
Speak the dream while the details are still fresh.
Transcribe
Whisper converts the recording into an editable text entry.
Interpret
A language model condenses the story into a visual prompt and summary.
Remember
The summary and generated image are stored together for later reflection.
Building a history, not a one-off image
Each summary and generated visual is stored in SQLite, allowing people to return to earlier entries and notice recurring themes over time. Streamlit kept the interface intentionally lightweight so recording, reviewing, and browsing could live in one simple flow.
Use AI to lower the effort of journaling while keeping the user’s own memory—not the generated interpretation—at the center.