The Year in AI Papers
The year/Independent research

Paper 2607.19064

Mage-Flow: An Efficient Native-Resolution Foundation Model for Image Generation and Editing

Xinjie Zhang, Peng Zhang, Shicheng Zheng, Jinghao Guo, Zhaoyang Jia, Yifei Shen, Xun Guo, Yuxuan Luo, Jiahao Li, Wenxuan Xie, Fanyi Pu, Xiaoyi Zhang, et al.

Published
Jul 2026
Research lab
Independent
Citations
1
GitHub
Not linked

01 In brief

Summary

Mage-Flow is a compact 4B-parameter generative stack from Microsoft for efficient text-to-image generation and instruction-based image editing.

It comprises Mage-VAE, a lightweight latent tokenizer using one-step diffusion-style encoding/decoding with anchor-latent regularization, and a Native-Resolution Multimodal Diffusion Transformer (NR-MMDiT) trained with rectified flow matching.

Mage-VAE reduces tokenization cost by over an order of magnitude while matching FLUX.2-VAE reconstruction quality.

Native-resolution packing and fused CUDA kernels enable flexible-resolution training and a 2.5x end-to-end training speedup.

The model family includes Base, RL-aligned (via Diffusion-NFT), and 4-step Turbo variants (via Decoupled-DMD with adversarial perceptual guidance).

At 1024x1024 on a single A100, Mage-Flow-Turbo generates an image in 0.59s and Mage-Flow-Edit-Turbo edits in 1.02s, with peak memory around 18-20 GB.

Despite its compact scale, Mage-Flow achieves competitive or superior performance on benchmarks like GenEval, DPG-Bench, and GEdit-Bench compared to much larger open-source models, demonstrating that careful co-design of tokenizer, backbone, and system can deliver strong high-resolution generation and editing within an efficient 4B model family.

02 From the paper

Abstract

Large-scale visual generators are increasingly capable but costly to train, fine-tune, and deploy. We introduce Mage-Flow, a compact 4B-scale generative stack for efficient text-to-image generation and instruction-based image editing. The stack is built from two co-designed components: Mage-VAE, a lightweight high-fidelity latent tokenizer, and a Native-Resolution Multimodal Diffusion Transformer trained with rectified flow matching. Mage-VAE uses one-step diffusion-style encoding and decoding with anchor-latent regularization, preserving the reconstruction quality of strong public VAEs while reducing tokenization cost by more than an order of magnitude. Together with native-resolution packing and stack-level CUDA kernel fusion, the stack supports flexible-resolution training and improves end-to-end training throughput by about $2.5\times$. Built on this foundation, we develop a complete model family with Base, RL-aligned, and Turbo variants for both generation and editing. Diffusion-NFT improves prompt following, text rendering, aesthetic quality, and editing fidelity, while few-step distillation with adversarial perceptual guidance produces 4-step Turbo models for low-latency inference. Despite its compact scale, Mage-Flow and Mage-Flow-Edit achieves competitive performance across standard generation and editing benchmarks. More importantly, the Turbo variants make high-resolution generation and editing practical for interactive use: at $1024^2$ resolution on a single NVIDIA A100 GPU, Mage-Flow-Turbo generates an image in 0.59s, and Mage-Flow-Edit-Turbo edits an image in 1.02s, while maintaining a small memory footprint. These results show that careful tokenizer--backbone--system co-design can deliver strong high-resolution generation and editing within an efficient 4B model family.