EXPLORINEER
en

arXiv 2611.07392

WeLM-Image: A Unified Image Generation Model towards Visual Intelligence

AI37 upvotes

Authors: Canmiao Fu, Chen Li, Guohui Zhang, Hao Lu, Hongyang Wei, Huadong Wu, Huaisong Zhang, Hubery Yin +8

Overview

The authors present WeLM-Image, a unified image foundation model that handles text-to-image generation, instruction-guided editing, and multi-reference synthesis in one interleaved sequence. It uses a Mixture-of-Transformers backbone with Mixture-of-Experts modules (31.7B stored parameters, about 5.4B active per target latent token), is built for single-GPU deployment in WeChat, and is distilled into a few-step Turbo variant with MeanFlow.

Introduction

Image generation, editing, and multi-reference synthesis are usually handled by separate systems, and large models are costly to deploy at consumer scale. WeLM-Image targets a single unified model that remains practical on one GPU.

Results

The 1536-resolution model ranks first among the evaluated open-source models on PRISM-Bench, Qwen-Image-Bench, and UnicBench. The MeanFlow-distilled WeLM-Image-Turbo generates a 1024×1024 image in 4.98 s on a single NVIDIA H20 GPU using eight guidance-free steps.

Conclusion

This shows a unified generation-and-editing model can reach top open-source benchmark rankings while staying fast enough for single-GPU consumer deployment.

Source: Read the paper on arXiv ↗

This explainer was rewritten in plain language by AI from the original article. Check the original for the facts.

Same story, different eyes