Feature Request: Add model selection and smart video trimming for talk videos

[DrKyro]

Feature Request

1. Model ID Selection

Allow users to choose their preferred model ID instead of using hardcoded defaults.

Current behavior:

• Qwen provider uses qwen3.5-flash by default
• OpenRouter uses stepfun/step-3.5-flash:free by default

Desired behavior:

• Add --model or --model-id CLI argument
• Allow users to specify any model ID (e.g., qwen-max, gpt-4o, etc.)
• In Streamlit UI, add a dropdown to select model

Example:

uv run python video_orchestrator.py --model-id qwen-max "VIDEO_URL"

2. Smart Video Trimming for Talk Videos (Smart Cut)

Automatically remove breath sounds (气口) and silence segments from talk videos to create tighter edits.

Desired behavior:

• Add --smart-cut or --trim-silence flag
• Detect silence segments (configurable threshold, e.g., -40dB)
• Detect breath sounds between sentences
• Automatically cut out these segments from the final clip
• Optionally keep a small buffer (e.g., 100ms) at cut points for natural transitions

Example:

uv run python video_orchestrator.py --smart-cut "VIDEO_URL"

Parameters to consider:

• --silence-threshold: dB threshold for silence detection (default: -40dB)
• --min-silence-duration: minimum silence duration to cut (default: 0.3s)
• --breath-detection: enable breath sound detection

---

Thanks for considering these features!

[linzzzzzz]

Thank you for the feature request. I agree they all make sense and will think about the planning (also feel free to take some if you want).

For model id selection, for now, if are comfortable with openrouter/qwen, you should be able to change model_id through default_model param at core/config.py.