v2.0.0

This major release introduces a two-pass subtitle generation workflow to significantly improve subtitle quality, accuracy, and refinement.

New Features & Improvements:

• Two-Pass Subtitle Generation: The core subtitle generation process has been re-architected into a two-pass system:
  • Pass 1 (Draft Generation): The first pass generates an initial draft of the subtitles. This is handled by the new SubtitlePass1JobRunner and configured with ai.pass1_model and thread.subtitles1.
  • Pass 2 (QA & Refinement): The second pass takes the JSON output from Pass 1 as a draft and performs quality assurance, correction, and refinement using a separate, specialized prompt. This is handled by the new SubtitlePass2JobRunner and configured with ai.pass2_model and thread.subtitles2.
• Improved Modularity: This new architecture allows for using different AI models for each pass (e.g., a faster model for the initial draft and a more powerful model for refinement).

Breaking Changes & Refactoring:

• Configuration (config.py):
  • The ai.model setting has been replaced by ai.pass1_model and ai.pass2_model. An optional ai.model shorthand is available to set both to the same value.
  • The thread.subtitles setting has been split into thread.subtitles1 and thread.subtitles2 to configure concurrency for each pass independently.
• Main Pipeline (main.py):
  • The main application logic has been significantly updated to orchestrate the new two-pass workflow, managing separate job queues and runners for each pass.
  • Job resumption logic now checks for the state of both passes to correctly resume interrupted sessions.
• Data Models (data_models.py):
  • SubtitleJob has been renamed to SubtitlePass1Job.
  • A new SubtitlePass2Job has been introduced to manage the state and data for the refinement pass.
• Agent Wrapper (agent_wrapper.py):
  • The RateLimitedAgentWrapper is now initialized with a specific model name, enabling the use of different models for each generation pass.