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

Researchers introduced WeLM-Image, a unified image foundation model designed for single-GPU deployment in WeChat consumer applications. The model integrates text-to-image generation, instruction-guided editing, and multi-reference synthesis using an interleaved sequence and a Mixture-of-Transformers backbone. In reported comparisons, the 1536-resolution model ranks first among evaluated open-source models on PRISM-Bench, Qwen-Image-Bench, and UnicBench, while a few-step distilled version generates a 1024x1024 image in 4.98 seconds. Open article →