Tagged articles

ReFM

1 articles · Page 1 of 1
ByteDance SE Lab
ByteDance SE Lab
Sep 21, 2026 · Artificial Intelligence

AI MediaKit: Multimodal Understanding & ReFM for Seamless Subtitle Removal

Volcano Engine's AI MediaKit uses multimodal semantic understanding to distinguish overlay subtitles from scene text, and a ReFM generative model with 4-step sampling, cross-frame context aggregation, and latent precise repasting to remove subtitles seamlessly while preserving original video quality, supporting 50+ languages and production-scale pipelines.

AI MediaKitReFMVolcano Engine
0 likes · 14 min read
AI MediaKit: Multimodal Understanding & ReFM for Seamless Subtitle Removal