models/README.md

# BitCheck Models

Place the Keras binary image detector at:

```text
models/deepfake_detector_efficientnetb0.keras
```

The current backend expects a Keras model whose output is a single real/authentic probability:

- `prob_real >= MODEL_THRESHOLD`: real/authentic
- `prob_real < MODEL_THRESHOLD`: AI-generated/synthetic

If this file is missing, BitCheck starts with `model_status: "missing"`. The rest of the verification pipeline still runs.