lrzjason/Anything2Real_demo_dataset

Overview

This dataset contains paired reference–target image sets, where each set includes:

The generation is used tori29umai/QwenImageEdit2509_LoRA/QIE_image2rakugaki_V1_dim4_1e-3-000012.safetensors to generate abstract image

A reference image, A corresponding text caption describing the from reference image to target image, A target image. It is designed for tasks such as image editing, style transfer, instruction-following image generation, or controlled image transformation.

Folder Structure & Naming Convention

Files are grouped into sets using the naming pattern: {basename}{idx}{suffix}.{ext}

{basename}: Shared identifier (e.g., 132, sceneA, etc.) grouping related sets. {idx}: Index within the group (e.g., 0, 1, …), allowing multiple variants per basename. {suffix}: Semantic role of the file: reference.png: Original/source image. reference.txt: Caption or prompt describing the reference image. target.png: Edited/modified/target image.

132_0_reference.png   # Reference image
132_0_reference.txt   # Caption for reference image
132_0_target.png      # Target/edit output

Intended Use Cases

Training/fine-tuning image editing models (e.g., Qwen-Image-Edit, LongCat-Image-Edit, etc.) Evaluating instruction-following fidelity in generative models. Studying semantic consistency between caption, source, and edit. Developing reference-guided generation or style transfer systems.