WAND-EraseVibe+ is a generative seamless erasure model. Its end-to-end latent-space architecture precisely strips subtitles and watermarks while reconstructing the frame — zero artifacts, frame-to-frame consistency, and flicker-free playback. It supports simultaneous multi-target erasure and processes over 100,000 minutes per day.

WAND-Vega is a family of multimodal application models built for vertical industries, spanning image generation, video generation, multimodal understanding and AI audio. Trained on Tencent-scale production data, it is tuned for commercial scenarios such as advertising, e-commerce, short comic dramas, embodied AI and AIGC tool platforms.

WAND-Codec+ is an encoding model driven by a VLM foundation model — the model understands the semantics of every frame and combines AI preprocessing for bitrate allocation and compression. It saves 50%+ bitrate versus open-source encoders, runs 2x+ faster, and supports 10+ mainstream codec standards.

WAND-Vega-VL takes unified video, audio, image and text input, delivering large-scale understanding for data annotation in embodied AI and autonomous driving. It also covers sports highlights, video summaries, product recognition and music performance review, with vertical models trained on real business data for sports, music and e-commerce.

WAND-Vega-Audio is an all-in-one audio AI model offering speech synthesis, voice cloning, vocal separation, and audio noise reduction. Spanning the full audio pipeline from generation and separation to enhancement and restoration, it supports AIGC voice restoration, multilingual dubbing and localization, and real-time AI commentary.