Faster text to image generation with one step adversarial training
ROAD: Modern Adversarial Training for Simple One-Step Distillation at Scale
Computer Vision and Pattern RecognitionMachine Learning
Summary
Generating images from text usually takes a lot of time and computing power. The authors introduce ROAD, a simpler way to speed up this process by training in just one step using a special technique called adversarial distillation. They improve the model with new components that better understand both images and text, and adjust training details to avoid common problems. This method achieves high-quality image results faster and with less computing effort than previous approaches.
What this means in practice
- •For machine learning engineers: Build faster text-to-image generators by distilling large pretrained models into efficient one-step models with fewer resources.
- •For ai service providers: Offer quicker image generation services on customer platforms by deploying lightweight models distilled with ROAD.$Commercial implications: This paper enables faster image generation products for clients needing quick and cost-effective AI art solutions.
Authors
Xiaoou Li, Shan You, Tao Huang
Abstract
One-step distillation turns pretrained diffusion models into fast generators, but strong results often require substantial training machinery and compute. Pure adversarial distillation offers a simpler route, yet has struggled at modern text-to-image scale. To address this limitation, we propose ROAD (Representation-grounded One-step Adversarial Distillation), a one-stage framework that distills text-to-image models with one adversarial objective and a lightweight representation-space critic. First, we replace the classical convolutional head with a Lipschitz-controlled Transformer head with token-level text conditioning, which matches the semantic tokens of modern visual encoders. Second, we observed a fixed random feature mixing leads to long-horizon degradation. We attribute it to generator-critic co-adaptation around one realization-specific view, and periodically re-draw the mixing to keep the randomization effective. Third, for our dense and vector-valued logits, we introduce Jacobian R1 which penalizes the full input-logit Jacobian, overcoming pooled R1's limitation of neglecting most of logit sensitivities. A single adversarial objective then suffices: in one step, ROAD reaches 11.03 FID on SDXL and 13.73 on SD3.5-M on COCO-10K, surpassing the multi-step teachers at a fraction of distribution matching's training compute.