Paper Detail
Ruikang Li, Zixuan Chen, Jiarui Wu, Mingde Yao, Yiyuan Zhang, Yujin Wang, Hanming Deng, Lewei Lu, Tianfan Xue
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 dynamics close to those of the base model and concentrate preservation corrections near the trajectory tail, where fine-grained deviations can be suppressed. These corrections are learned from the editor's own on-policy denoising rollouts, 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.
No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.
No ranking explanation is available yet.
No tags.
@misc{li2026tailsteer,
title = {TailSteer: Trajectory-Tail Steering for Drift-Resistant Multi-turn Image Editing},
author = {Ruikang Li and Zixuan Chen and Jiarui Wu and Mingde Yao and Yiyuan Zhang and Yujin Wang and Hanming Deng and Lewei Lu and Tianfan Xue},
year = {2026},
abstract = {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 i},
url = {https://huggingface.co/papers/2611.06741},
keywords = {multi-turn image editing, visual drift, cross-turn rollouts, pretrained priors, inference-time mechanisms, foundation models, final-stage training, denoising dynamics, trajectory tail, on-policy denoising rollouts, preservation reference, ground-truth editing pairs, cumulative drift, text-to-image, single-image editing, long-horizon editing robustness, FLUX, SenseNova, Qwen, code available, huggingface daily},
eprint = {2611.06741},
archiveprefix = {arXiv},
}
{}