a1557811266/

RAD_DataSet

A large-scale, fully synthetic dataset for image editing tasks, containing 514,510 high-quality annotated image tuples.

The RAD dataset is generated through a rigorous three-stage pipeline:

Textual Prompt Generation: Structured guidance prompts generated using Qwen3-32B LLM Image Pair Synthesis: Diptych image pairs synthesized using FLUX.1 diffusion transformer, fine-tuned evolutionarily via LoRA Quality Filtering &… See the full description on the dataset page: https://huggingface.co/datasets/a1557811266/RAD_DataSet.