Skip to content

v2.6.1b1

Pre-release
Pre-release

Choose a tag to compare

View all tags
@FlippFuzz FlippFuzz released this 29 Mar 11:35
v2.6.1b1
4b4fe67

This release focuses on hardening the AI's transcription logic and translation accuracy by introducing stricter boundaries for reference data and expanding the contextual window for translations.

Prompt Engineering:

  • Subtitle Generation (v14):
    • The Manual Transcription Mandate: Explicitly instructs the AI that it is "not exempt" from subtitling if the reference JSON is null or mismatching; it must rely on native audio perception to transcribe.
    • Strict Scene Boundaries: Added rules to prevent "cross-contamination" where lyrics from a later scene in the reference JSON are incorrectly applied to earlier audio.
    • Enhanced Contextual Window: Refined the translation logic to mandate analysis of the Previous 2 Sentences and Next 2 Sentences to improve pronoun resolution and tonal consistency.
    • Refined Anti-Hallucination: Updated the "Wrong Song" and "Partial Video" scenarios with specific timestamp-based examples to help the model ignore irrelevant reference data.
    • Improved Decoding Hierarchy: Streamlined the fallback logic for resolving ambiguous audio, prioritizing visual OCR and scene context.

Full Changelog: 2.6.0...v2.6.1b1

Assets 2
Loading