Paper 2605.13565
Qwen-Image-VAE-2.0 Technical Report
01 In brief
Summary
Qwen-Image-VAE-2.0 is a suite of high-compression image VAEs (f16 and f32) designed to overcome the trade-off between compression ratio, reconstruction fidelity, and diffusability.
The architecture uses Global Skip Connections (GSC) to preserve fine details, expanded latent channels, and an attention-free, asymmetric encoder-decoder backbone for efficiency.
Training scales to billions of images, includes a synthetic document-rendering pipeline for text-rich scenarios, and removes KL and GAN losses.
A semantic alignment strategy using DINOv2 middle-layer features improves latent diffusability.
The new OmniDoc-TokenBench benchmark (~3K text-rich images) evaluates text fidelity via OCR-based Normalized Edit Distance (NED).
Results show state-of-the-art reconstruction on ImageNet and FFHQ, and on OmniDoc-TokenBench, Qwen-Image-VAE-2.0-f16c128 achieves NED 0.9617, surpassing f8 baselines, while f32c192 reaches 0.8555.
Downstream SiT experiments demonstrate faster convergence and better generation quality than existing high-compression VAEs.
The models are integrated into Qwen-Image-2.0, validating scalability.
02 From the paper
Abstract
We present Qwen-Image-VAE-2.0, a suite of high-compression Variational Autoencoders (VAEs) that achieve significant advances in both reconstruction fidelity and diffusability. To address the reconstruction bottlenecks of high compression, we adopt an improved architecture featuring Global Skip Connections (GSC) and expanded latent channels. Moreover, we scale training to billions of images and incorporate a synthetic rendering engine to improve performance in text-rich scenarios. To tackle the convergence challenges of high-dimensional latent space, we implement an enhanced semantic alignment strategy to make the latent space highly amenable to diffusion modeling. To optimize computational efficiency, we leverage an asymmetric and attention-free encoder-decoder backbone to minimize encoding overhead. We present a comprehensive evaluation of Qwen-Image-VAE-2.0 on public reconstruction benchmarks. To evaluate performance in text-rich scenarios, we propose OmniDoc-TokenBench, a new benchmark comprising a diverse collection of real-world documents coupled with specialized OCR-based evaluation metrics. Qwen-Image-VAE-2.0 achieves state-of-the-art reconstruction performance, demonstrating exceptional capabilities in both general domains and text-rich scenarios at high compression ratio. Furthermore, downstream DiT experiments reveal our models possess superior diffusability, significantly accelerating convergence compared to existing high-compression baselines. These establish Qwen-Image-VAE-2.0 as a leading model with high compression, superior reconstruction, and exceptional diffusability.