RL Post-Training in Diffusion Transformers Is Activation Steering
Tuna Han Salih Meral, Kaan Oktay, Alec Helbling, Atakan Kara, Duen Horng Chau, Pinar Yanardag
Abstract
Reinforcement learning (RL) post-training substantially improves the visual quality, text rendering, and compositional accuracy of diffusion transformers. Because capabilities such as rendering requested text or composing multiple objects are inherently prompt-dependent, a natural hypothesis is that improving them requires learning complex, input-dependent transformations. To isolate how much of these gains an input-independent shift explains, we probe each RL-tuned model with a deliberately simple parameterization: a fixed set of steering vectors added to the block outputs of the frozen base model. On SD3.5-M, these vectors hold 65x fewer parameters than the RL-tuned LoRA (0.29M against 18.8M). We fit them to match the velocity predictions of public RL-tuned checkpoints, with no reward in the objective. Although the vectors are identical for every prompt, token, and sampling seed, they recover 90.0-96.3% of the reward gains of six RL-tuned image models, and 89.8% and 111.9% on two RL-tuned video models. Initialized at zero and trained with online RL, the vectors alone exceed the reward gains of weight-based RL on text rendering (OCR) and human preference (PickScore) on SD3.5-M. A constant shift thus suffices for most of the gain of policy-gradient RL post-training: without new input-dependent transformations, it steers the pretrained generator toward what it can already compute.
Create a lesson
Related papers
Posterior Matching Distillation
Hidir Yesiltepe, Julius Berner, Arash Vahdat et al.
One Prior, Two Dynamics: Repurposing Video Diffusion Transformers for Self-Supervised Latent Action Learning
Yuxin Jiang, Wenqiang Sun, Jiaqi Li et al.
ScaleWorld: High-Resolution World Generation through Shared Prediction and Refinement
Yansong Qu, Wesker Zheng, Quanjian Song et al.
Depth Pro 2: Sharp and Temporally Consistent Metric Monocular Depth
Peiyun Hu, Marko Mihajlovic, Lyndon Duong et al.
Coda-DaVinci: Reimagining Video Generation with Code2Video
Yiyan Ji, Qixun Wang, Yang Shi et al.
Multi-institutional grading of Barrett's esophagus dysplasia without stain normalization using a pathology foundation model and multiple instance learning
Amirali Khosravi, Jaidip M. Jagtap, D. Chamil Codipilly et al.