TailSteer: Trajectory-Tail Steering for Drift-Resistant Multi-turn Image Editing

Strong single-turn editing performance does not guarantee reliable multi-turn interaction: as edited outputs are repeatedly reused, small unintended changes can accumulate into severe visual drift. This long-term drift is often mitigated through training on multi-turn data or cross-turn rollouts, potentially altering pretrained priors, or through specialized inference-time mechanisms that are difficult to scale to foundation models. We introduce TailSteer, a final-stage training framework that improves multi-turn robustness while retaining the generative and editing capabilities acquired through earlier training. TailSteer follows a simple principle: keep early denoising dynamicsdenoising dynamicsThe step‑by‑step process a diffusion model uses to remove noise and generate an image. close to those of the base model and concentrate preservation corrections near the trajectory tailtrajectory tailThe later stages of the diffusion denoising process where fine details are refined., where fine-grained deviations can be suppressed. These corrections are learned from the editor's own on-policy denoising rolloutson-policy denoising rolloutsGenerated denoising sequences using the current editor model as it would be used at inference time., using unchanged source content as a preservation reference without requiring curated ground-truth editing pairs. Experiments across the FLUX, SenseNova, and Qwen families show that TailSteer substantially reduces cumulative drift over 15-20 editing turns while largely preserving base-model performance on extensive text-to-image and single-image editing benchmarks. These results demonstrate that long-horizon editing robustness can be unlocked in foundation models after TailSteer, rather than by relearning their existing capabilities.