Uni MoE

Uni MoE is a family of multimodal AI models for researchers working with language, visual material and audio together. Its Omni model can reason across combined inputs and answer questions about their content.

• Visual and audio understanding: Process images and video alongside text, speech, music and environmental sounds, including math questions presented in images.
• Speech and text responses: The Omni model generates written answers and speech that uses conversational context, and handles long speech recordings for understanding and synthesis.
• Image creation and editing: Generate pictures from text or use instructions to remove objects, change colors, add items and reduce rain or haze in an image.
• Guided image generation: Use edge maps or depth maps to guide the structure of a generated picture.
• Voice and music generation: The Uni-MoE-Audio branch supports voice cloning, text-to-speech, and music generation from text or video.

The Omni model is built on Qwen2.5-7B. Uni-MoE-Audio is a separate branch of the family with its own audio generation capabilities.