High-quality training triplets (source-target image pairs with precise editing instructions) are a critical bottleneck for scaling instruction-guided image editing models. Vision-language models (VLMs) are widely used for automated instruction synthesis, but we identify three systematic failure modes in image-pair settings: orientation inconsistency (e.g., left/right confusion), viewpoint ambiguity, and insufficient fine-grained attribute description. Human evaluation shows that over 47% of instructions from strong baseline VLMs contain critical errors unusable for downstream training. We propose EditCaption, a scalable two-stage post-training pipeline for VLM-based instruction synthesis. Stage 1 builds a 100K supervised fine-tuning (SFT) dataset by combining GLM automatic annotation, EditScore-based filtering, and human refinement for spatial, directional, and attribute-level accuracy. Stage 2 collects 10K human preference pairs targeting the three failure modes and applies direct preference optimization (DPO) for alignment beyond SFT alone. On Eval-400, ByteMorph-Bench, and HQ-Edit, fine-tuned Qwen3-VL models outperform open-source baselines; the 235B model reaches 4.712 on Eval-400 (vs. Gemini-3-Pro 4.706, GPT-4.1 4.220, Kimi-K2.5 4.111) and 4.588 on ByteMorph-Bench (vs. Gemini-3-Pro 4.522, GPT-4.1 3.412). Human evaluation shows critical errors falling from 47.75% to 23% and correctness rising from 41.75% to 66%. The work offers a practical path to scalable, human-aligned instruction synthesis for image editing data.

翻译：高质量的“三元组”（即包含精准编辑指令的源-目标图像对）是扩展指令引导型图像编辑模型的关键瓶颈。尽管视觉-语言模型（VLM）被广泛用于自动化指令合成，但我们在图像对场景中识别出三种系统性失败模式：方向不一致（如左右混淆）、视角模糊性以及细粒度属性描述不足。人工评估表明，超过47%由强基线VLM生成的指令存在严重错误，无法用于下游训练。为此，我们提出EditCaption——一种可扩展的两阶段VLM指令合成后训练流水线。阶段一构建包含10万条监督微调（SFT）样本的数据集，融合GLM自动标注、基于编辑分数的过滤以及人工精调，确保空间、方向和属性层面的准确性。阶段二收集针对三种失败模式的1万个人类偏好对，并应用直接偏好优化（DPO）实现超越SFT的对齐。在Eval-400、ByteMorph-Bench和HQ-Edit基准上，微调后的Qwen3-VL模型超越开源基线；其中235B模型在Eval-400上达到4.712（对比Gemini-3-Pro 4.706、GPT-4.1 4.220、Kimi-K2.5 4.111），在ByteMorph-Bench上达到4.588（对比Gemini-3-Pro 4.522、GPT-4.1 3.412）。人工评估显示，严重错误率从47.75%降至23%，正确率从41.75%提升至66%。该工作为图像编辑数据提供了一条可扩展且人类对齐的指令合成实用路径。