ChordEdit: One-Step Low-Energy Transport
for Image Editing

Liangsi Lu¹, Xuhang Chen², Minzhe Guo¹, Shichu Li³, Jingchao Wang⁴, Yang Shi^1†
¹ Guangdong University of Technology, ² Huizhou University, ³ Shenzhen University, ⁴ Peking University
^† Corresponding author

Paper Code

Figure 1. ChordEdit Capabilities. Our model-agnostic, training-free, and inversion-free method facilitates real-time editing on fast generative models.

Abstract

The advent of one-step text-to-image (T2I) models offers unprecedented synthesis speed. However, their application to text-guided image editing remains severely hampered, as forcing existing training-free editors into a single inference step fails. This failure manifests as severe object distortion and a critical loss of consistency in non-edited regions, resulting from the high-energy, erratic trajectories produced by naive vector arithmetic on the models' structured fields. To address this problem, we introduce ChordEdit, a model agnostic, training-free, and inversion-free method that facilitates high-fidelity one-step editing. We recast editing as a transport problem between the source and target distributions defined by the source and target text prompts. Leveraging dynamic optimal transport theory, we derive a principled, low-energy control strategy. This strategy yields a smoothed, variance-reduced editing field that is inherently stable, facilitating the field to be traversed in a single, large integration step. A theoretically grounded and experimentally validated approach allows ChordEdit to deliver fast, lightweight and precise edits, finally achieving true real-time editing on these challenging models.

Methodology

Existing training-free methods fail in one-step settings because the naive drift difference creates a high-energy, erratic trajectory. When integrated in a single step, this leads to significant error accumulation.

Figure 2. Comparison of Editing Field Stability. (Left) Naive one-step simple drift is high-energy and volatile. (Right) ChordEdit derives a stable, low-energy Chord Control Field by time-averaging the observable fields, facilitating accurate single-step transport.

Chord Control Field

Our key contribution is the Chord Control Field, a theoretically-grounded, time-weighted average of the source and target drifts. This acts as a temporal smoothing operator, yielding an inherently stable, low-energy field.

Results

Qualitative Comparison

We compare ChordEdit with state-of-the-art multi-step, few-step, and one-step editors. ChordEdit consistently adheres to the prompt with exceptional background preservation, avoiding artifacts or identity failures seen in other methods.

Figure 3. Qualitative Comparison on PIE-bench. ChordEdit demonstrates a strong balance of semantics and consistency compared to other few-step and multi-step methods.

Ablation Study: Energy Analysis

To validate the stability of our method, we visualize the energy of the editing fields. The Naive baseline exhibits high-energy spikes, leading to severe artifacts and background corruption. In contrast, ChordEdit derives a stable, low-energy field, resulting in high-fidelity edits that preserve object identity and non-edited regions.

Figure 4. Ablation Study and Energy Visualization. We compare ChordEdit (Ours) against the naive baseline. The color maps show the energy of the editing field. Note how the naive method's high energy correlates with destroyed backgrounds (e.g., in the "dog -> wolf" example), whereas ChordEdit maintains low energy and clean backgrounds .

BibTeX


@misc{lu2026chordedit,
      title={ChordEdit: One-Step Low-Energy Transport for Image Editing}, 
      author={Liangsi Lu and Xuhang Chen and Minzhe Guo and Shichu Li and Jingchao Wang and Yang Shi},
      year={2026},
      eprint={2602.19083},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2602.19083}, 
}

ChordEdit: One-Step Low-Energy Transportfor Image Editing