OmniAgent: Native Active Perception as Reasoning for Omni-Modal Understanding
An agentic audio-visual understanding model that iteratively requests frames, audio, and clips to answer questions — powered by a Qwen2.5-Omni-7B model fine-tuned with agentic SFT and agentic RL.
🤗 Huge thanks to the Hugging Face team for building this interactive demo based on our original UI! Please note that as an adapted version, its underlying logic is not 100% identical to our official release. For the exact standard and fully-featured experience, please use our official demo script.
Examples