# Microsoft Open-Sources Lightweight Multimodal Image Generation Model Series Mage-Flow

- Category: Flash (快讯)
- Published: 2026-07-23T11:56:00.000Z
- Source date: 2026-07-23
- HTML: https://www.88z.ai/en/news/flash-20260723-bb8768683ca59b4d
- Markdown: https://www.88z.ai/en/news/flash-20260723-bb8768683ca59b4d.md
- Other language: https://www.88z.ai/news/flash-20260723-bb8768683ca59b4d.md

Microsoft has released Mage-Flow, an open-source series of lightweight multimodal image generation models based on flow matching architecture with only 4 billion parameters. The series includes Base, RL, Turbo, and Edit versions, supporting text-to-image generation, local editing, and style transfer. It runs efficiently on consumer-grade GPUs, suitable for e-commerce, advertising, and game development.

## At a glance

- Mage-Flow uses flow matching architecture with 4B parameters across four versions
- Runs on consumer GPUs, suitable for e-commerce and game development

## The update

Microsoft has officially open-sourced Mage-Flow, a series of lightweight multimodal image generation models. Built on a flow matching architecture, the model keeps its parameter count to 4 billion, significantly reducing computational resource requirements. The Mage-Flow series offers four distinct versions: Base, RL, Turbo, and Edit, catering to diverse application needs. Functionally, it supports text-to-image generation, local image editing, and style transfer tasks. Thanks to its lightweight design, the model runs efficiently on consumer-grade GPUs, substantially lowering the barrier to deployment. This makes it particularly suitable for scenarios sensitive to efficiency and cost, such as e-commerce design, advertising creative production, and game development, providing developers with a more accessible image generation solution.

## Sources

- [Mage-Flow](https://www.88z.ai/flash/cite/flash-20260723-bb8768683ca59b4d) (2026-07-23)

This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.