[GenRamzi]

Improvements

• Detect and select installed Ollama models dynamically.
• Add configurable shot duration and rotate uploaded media through the full narration.
• Add subtitle font size, text color, outline color, outline width, and position controls.
• Add neural voice preview from the studio before rendering.
• Expand Arabic voice choices.
• Clean temporary voice-preview files after delivery.
• Add provider/security documentation and clarify the deliberate NoAPI stock-media tradeoff.
• Add tests for new schema validation and subtitle color conversion helpers.

Design

Cloud account providers continue to use official vendor CLIs. Local open-weight model families are exposed through Ollama without hard-coding a fake web-login mechanism for each model vendor.

Validation

CI compiles the Python package and runs pytest. This PR is draft until those checks pass.