Research
TailSteer proposes a training method to reduce drift in multi-turn image editing
TailSteer is a final-stage training framework that keeps early denoising close to the base model and corrects deviations near the trajectory tail. Tested on FLUX, SenseNova and Qwen models, it reportedly cuts cumulative drift over 15-20 editing turns while largely preserving base performance.
Every outlet that ran it 1 article
Hugging Face PapersResearchLead
TailSteer: Trajectory-Tail Steering for Drift-Resistant Multi-turn Image Editing