The Year in AI Papers
The year/Independent research

Paper 2508.03320

Skywork UniPic: Unified Autoregressive Modeling for Visual Understanding and Generation

Peiyu Wang, Yi Peng, Yimeng Gan, Liang Hu, Tianyidan Xie, Xiaokun Wang, Yichen Wei, Chuanxin Tang, Bo Zhu, Changshi Li, Hongyang Wei, Eric Li, et al.

Published
Aug 2025
Research lab
Independent
Citations
17
GitHub
871 stars

01 In brief

Summary

Skywork UniPic is a 1.5B-parameter autoregressive model that unifies image understanding, text-to-image generation, and image editing in a single architecture without task-specific adapters.

It achieves a GenEval score of 0.86, a DPG-Bench score of 85.5, and GEditBench-EN and ImgEdit-Bench scores of 5.83 and 3.49, respectively, while generating 1024x1024 images on commodity hardware (e.g., RTX 4090).

The architecture uses a decoupled encoding strategy: a Masked Autoregressive (MAR) encoder for generation and a SigLIP2 encoder for understanding, both feeding a shared Qwen2.5-1.5B language model backbone.

Training involves a four-stage progressive curriculum scaling resolution from 256x256 to 1024x1024, with dynamic parameter unfreezing.

Data quality is ensured via two reward models (Skywork-ImgReward and Skywork-EditReward) trained with GRPO and SFT, filtering samples below a 0.9 quality threshold.

The model demonstrates that compact unified multimodal systems can achieve state-of-the-art performance with about one-tenth the parameters of larger models like BAGEL (14B) or UniWorld-V1 (19B).

Limitations include occasional struggles with complex instructions and incomplete edits.

The authors open-source the model weights and code.

02 From the paper

Abstract

We introduce Skywork UniPic, a 1.5 billion-parameter autoregressive model that unifies image understanding, text-to-image generation, and image editing within a single architecture-eliminating the need for task-specific adapters or inter-module connectors-and demonstrate that compact multimodal systems can achieve state-of-the-art performance on commodity hardware. Skywork UniPic achieves a GenEval score of 0.86, surpassing most existing unified models; sets a new DPG-Bench complex-generation record of 85.5; attains 5.83 on GEditBench-EN and 3.49 on ImgEdit-Bench for image editing; and generates 1024 x 1024 images with under 15 GB of GPU memory (e.g., RTX 4090). (1) a decoupled encoding strategy that leverages a masked autoregressive encoder for synthesis and a SigLIP2 encoder for understanding, all feeding a shared autoregressive decoder; (2) a progressive, resolution-aware training schedule scaling from 256 x 256 to 1024 x 1024 while dynamically unfreezing parameters to balance capacity and stability; and (3) meticulously curated, 100 million-scale datasets augmented with task-specific reward models to refine generation and editing objectives. By demonstrating that high-fidelity multimodal integration need not incur prohibitive resource demands, Skywork UniPic establishes a practical paradigm for deployable, high-fidelity multimodal AI. Code and weights are publicly available at https://huggingface.co/Skywork/Skywork-UniPic-1.5B.