# AI Video Model Comparison Dashboard (Streamlit + Plotly)

## PURPOSE
An interactive Streamlit dashboard that visualises the head-to-head test results
from the video: Kling 4.0 versus Seedance 2.5 across cost, physics realism, lip-
sync, character consistency and text-to-video, with the cheaper LTX 2.5 shown as
the budget tier. It turns the host's side-by-side review into charts a team can
filter — cost bars, a capability radar, a score-vs-price scatter, and a model
feature matrix.

## WHY THIS MATTERS
The transcript's verdict is nuanced, not binary: Kling 4 is ~70% cheaper
(~$0.80 vs ~$2.85 for a 5-second 1080p clip), reaches 30 seconds, uses up to 10
key frames, and adds 10-bit, 21:9, HDR, 8,000-token prompts and German/French/
Hindi — yet Seedance 2.5 still wins the physics tests (more random glass
shattering, varied droplets) and remains the host's preferred pick for specific
sequences. A dashboard makes that trade-off visible instead of a takeaway line.

## HOW TO BUILD
1. `pip install streamlit plotly pandas`
2. Build small Pandas frames: a `cost_df`, a `score_df` per test category, and a
   `features_df` of booleans (30s, 10 keyframes, HDR, 21:9, multi-language).
3. Render a Plotly grouped bar for cost, a `go.Scatterpolar` radar for quality,
   and `st.dataframe` for the feature matrix.
4. Add a model multiselect so reviewers can focus on one tier.

```python
fig = go.Figure()
fig.add_trace(go.Scatterpolar(
    r=radar["kling-4.0"], theta=CATEGORIES, fill="toself", name="kling-4.0"))
fig.add_trace(go.Scatterpolar(
    r=radar["seedance-2.5"], theta=CATEGORIES, fill="toself", name="seedance-2.5"))
st.plotly_chart(fig, use_container_width=True)
```

## WHEN TO USE IN PRODUCTION
Use it in model-selection reviews, to justify a subscription tier to stakeholders,
or as the "review website" the video describes where generated clips and their
prompts are collected for approval. Not suited to real-time per-job monitoring,
and it should not be the sole basis for a creative decision a human should judge.

## HOW IT HELPS STUDENTS
Students practise Streamlit state and layout, Plotly chart construction from tidy
data, and communicating a multi-dimensional comparison honestly — including
showing where the cheaper model loses.

## STARTER CODE
`starter/main.py` is a runnable Streamlit app with bundled transcript-derived
data, three charts, and `# TODO` markers for loading real benchmark CSVs.
