🖼 Microsoft Unveils Mage-Flow for Fast Image Generation
The company has released a family of compact multimodal models with 4 billion parameters. Thanks to the innovative Mage-VAE tokenizer, encoding efficiency has increased by 12–22 times. The Turbo version, running on NVIDIA A100, generates images in just 0.59 seconds.
🌍 Mage-Flow proves that compact models (4B) can compete with giants (20B+) in editing and generation tasks through architectural efficiency, reducing resource requirements.
👤 This allows high-quality image creation and deep editing tools to be run on accessible hardware with near-instant response times.
Source 1: https://microsoft.github.io/Mage/flow/ Source 2: https://github.com/microsoft/Mage