MT Forcing: Towards Stable and Consistent Multi-Turn Image Editing with Autoregressive Diffusion
Paper · https://arxiv.org/abs/2610.25404
Multi-turn image editing aims to modify an image according to a sequence of instructions, enabling more effective interactive visual content creation for users. However, existing methods often exhibit noticeable drift after only a few editing turns, manifesting in two aspects: 1) unstable visual quality, with accumulating artifacts and degraded details; and 2) inconsistent image content, with unintended changes to preserved regions and prior edits. To address these challenges, we propose MT Forcing, a novel training paradigm for multi-turn image editing with unified multimodal models (UMMs) based on autoregressive diffusion. Specifically, we propose Single-Turn Distribution Matching Distillation (Single-Turn DMD), which supervises student self-rollouts with a teacher that applies cumulative edits to the source image in a single turn, providing visual-quality guidance without recursive error accumulation. Additionally, we introduce Teacher-Forcing Guidance, which supervises history-dependent transitions on reference trajectories during warmup and as a regularizer during distillation, encouraging cross-turn content consistency beyond what per-turn distribution matching alone can ensure. Extensive experiments demonstrate that our approach significantly improves the multi-turn image editing capabilities of UMMs, yielding more stable visual quality and stronger cross-turn consistency.
First seen by Oriel on .
AI summary
MT Forcing is a training paradigm for autoregressive diffusion models that improves stability and consistency in multi-turn image editing.
- Multi-turn editing often accumulates artifacts, degrades details, and unintentionally changes preserved regions or earlier edits.
- Single-Turn Distribution Matching Distillation supervises student self-rollouts using a teacher that applies cumulative edits to the source image in one turn.
- Teacher-Forcing Guidance trains history-dependent transitions on reference trajectories during warmup and regularizes distillation for cross-turn consistency.
- Experiments report significantly stronger multi-turn image editing capabilities, with more stable visual quality and improved consistency across turns.
#image editing #autoregressive diffusion #multimodal models
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Hugging Face Papers: 36, latest observation ; +2 over observed period since
Comments
Comment count reported by the source.
- Hugging Face Papers: 1, latest observation ; 0 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Papers | 2610.25404 | https://arxiv.org/abs/2610.25404 Read discussion |