README.md

---
title: Station Vision TTS
emoji: 🚉
colorFrom: blue
colorTo: green
sdk: gradio
sdk_version: 4.44.1
app_file: app.py
pinned: false
---

# Station Vision TTS Space

Remote Gradio demo for the Japanese station-announcement pipeline:

1. `LiquidAI/LFM2.5-VL-450M-Extract` extracts compact JSON from an image.
2. `LiquidAI/LFM2-350M` turns the JSON into one short Japanese announcement.
3. `3bitquantizers/lfm25-audio-jp-station-yamanote-merged-v1-all-langs-static` synthesizes live audio.

## Deploy

Create a GPU-backed Gradio Space under `3bitquantizers`, then copy these files to the Space root:

```bash
cp app.py requirements.txt README.md /path/to/space/
cp ../image_to_station_audio.py /path/to/space/
```

Set Space secrets:

- `HF_TOKEN`: read access to private `3bitquantizers` model repos.

Optional Space variables:

- `VL_MODEL_ID`
- `PROMPT_MODEL_ID`
- `TTS_MODEL_ID`
- `MAX_TTS_TOKENS`

Use a 24 GB GPU tier or larger for the first smoke. The app keeps models warm and queues requests one at a time.