FireRed-Image-Edit-1.0 Technical Report

TL;DR

FireRed-Image-Edit introduces a diffusion transformer for instruction-based image editing, achieving state-of-the-art results through extensive data curation, multi-stage training, and innovative optimization techniques, supported by a new comprehensive benchmark.

Contribution

The paper presents a novel diffusion transformer model with advanced training strategies and a large, high-quality dataset for instruction-based image editing, along with a new benchmark suite.

Findings

Achieves state-of-the-art performance on REDEdit-Bench and public benchmarks.

Introduces new techniques like Multi-Condition Aware Bucket Sampler and Asymmetric Gradient Optimization.

Demonstrates strong semantic coverage and instruction alignment in image editing tasks.

Abstract

We present FireRed-Image-Edit, a diffusion transformer for instruction-based image editing that achieves state-of-the-art performance through systematic optimization of data curation, training methodology, and evaluation design. We construct a 1.6B-sample training corpus, comprising 900M text-to-image and 700M image editing pairs from diverse sources. After rigorous cleaning, stratification, auto-labeling, and two-stage filtering, we retain over 100M high-quality samples balanced between generation and editing, ensuring strong semantic coverage and instruction alignment. Our multi-stage training pipeline progressively builds editing capability via pre-training, supervised fine-tuning, and reinforcement learning. To improve data efficiency, we introduce a Multi-Condition Aware Bucket Sampler for variable-resolution batching and Stochastic Instruction Alignment with dynamic prompt re-indexing. To stabilize optimization and enhance controllability, we propose Asymmetric Gradient Optimization for DPO, DiffusionNFT with layout-aware OCR rewards for text editing, and a differentiable Consistency Loss for identity preservation. We further establish REDEdit-Bench, a comprehensive benchmark spanning 15 editing categories, including newly introduced beautification and low-level enhancement tasks. Extensive experiments on REDEdit-Bench and public benchmarks (ImgEdit and GEdit) demonstrate competitive or superior performance against both open-source and proprietary systems. We release code, models, and the benchmark suite to support future research.