AI Dataset Cropper

Prepare image datasets with consistent crops

Training images work best when their dimensions and framing are consistent. ImageCropKit gives you a browser-based workflow for cropping many dataset images at once. Upload a batch, choose a training size, review the crops, and export everything as a ZIP.

Common AI training image sizes

Different training workflows use different image sizes. ImageCropKit includes the sizes most often needed for square, portrait, and landscape dataset preparation:

• 512 × 512 for lightweight SD 1.5-style datasets
• 768 × 768 for mid-resolution square training images
• 1024 × 1024 for SDXL-style square datasets
• 1024 × 1536 for portrait images
• 1536 × 1024 for landscape images

LoRA, Stable Diffusion, and Flux workflows

This page is a general AI dataset cropper. If you are specifically preparing a LoRA dataset, the LoRA image cropper provides a focused workflow for common LoRA sizes. For square-only datasets, the 1:1 cropper is useful when you only need one image.

What this tool does not do

ImageCropKit does not judge dataset quality, remove duplicates, caption images, or train models. It focuses on one preparation step: creating consistent crops and exact output dimensions before you move images into your training pipeline.

Private dataset preparation

Dataset images can contain unreleased characters, client work, research material, or private references. ImageCropKit processes images locally in the browser, so files are not uploaded while you crop and export the dataset.