Swift-Image: Pushing Compact Unified Image Generation and Editing to the Performance Frontier

Forum topic · 小凯 · 2026-08-22

Summary

Swift-Image is a compact unified model for text-to-image generation, single-image editing, and multi-image editing, presented in arXiv paper 2608.20334 by Taihang Hu, Zhao Wang, Zuan Gao, and Tao Liu. The work investigates how far a relatively small visual generator can be pushed under a constrained compute budget through systematic training engineering. Swift-Image uses an efficient 6B single-stream Diffusion Transformer (DiT) with a progressive training pipeline that evolves from broad semantic coverage toward higher resolution, stronger visual quality, and unified generation-editing supervision. Post-training combines parallel expert reinforcement learning with multi-teacher on-policy distillation to mitigate interference between heterogeneous objectives, and a Prompt Enhancer translates user requests into generator-aligned visual specifications, decoupling high-level reasoning from pixel-level rendering. Structural pruning and few-step distillation yield 3B and accelerated variants. The 6B model achieves leading overall performance among evaluated open-source models using only 243K GPU training hours, with the 3B compressed variant showing nearly no loss and few-step distillation further improving editing performance with far fewer sampling steps.

Paper Overview

Field: Computer Vision (CV) Authors: Taihang Hu, Zhao Wang, Zuan Gao, Tao Liu Published: 2026-08-22 arXiv: 2608.20334

Abstract

Swift-Image is a compact unified model supporting text-to-image generation, single-image editing, and multi-image editing. The central question of the work: how far can a relatively small visual generator be pushed under a constrained compute budget, through systematic training engineering?

Key Technical Points

Architecture: An efficient 6B single-stream DiT (Diffusion Transformer) as the unified visual generator.
Progressive training pipeline: Training evolves from broad semantic coverage toward higher resolution, stronger visual quality, and unified generation-editing supervision.
Post-training: Parallel expert reinforcement learning followed by multi-teacher on-policy distillation, designed to mitigate interference between heterogeneous objectives.
Prompt Enhancer: Translates user requests into generator-aligned visual specifications, decoupling high-level reasoning from pixel-level rendering.
Compression: Structural pruning and few-step distillation produce 3B and accelerated variants.

Results

With only 6B parameters and 243K GPU training hours, Swift-Image achieves leading comprehensive performance among evaluated open-source models.
The compressed 3B model retains nearly all performance.
Few-step distillation further improves overall editing performance while substantially reducing sampling steps.

---

*Auto-collected on 2026-08-22.*