dailyai.report

Researchers at Apple developed a new framework to fix modal interference in text-to-sounding video generation. The system addresses the gap between dense training captions and short user prompts. By refining cross-modal feature interaction, the approach ensures tighter synchronization between audio and visuals. This reduces the common misalignment seen in joint-training models.