I made this repo to easily caption and organize images for training AI image models. It supports uploading images (PNG/JPG) or videos (MP4) to auto generate images at 1 second intervals. You can add captions to each image and download a ZIP file with all images and associated captions in txt files.
I use it specifically for training Stable Diffusion models and creating a training set in the format used by kohya_ss. Note: this repo is NOT associated in any way with Stable Diffusion nor kohya_ss.
Privacy is a pivotal aspect of this repo. Your data will never be sent outside your computer. No data will be saved in your browser either; if you refresh or close a tab, any working data will be lost.
- Prerequisite:
NodeJS(I use v20.11.1, but other versions probably work too) npm installnpm run devto run on port 3000
- Upload one or multiple images (PNG/JPG) and add a caption for each
- The more detailed the captions, the better the training set
- Upload an MP4 video file
- Set a start and end time to capture images from the video
- Images can be generated at 1 second intervals
- Or take a screenshot of particular frames
- Add a caption for generated images
- Download a ZIP file with all images and associated captions in txt files
- Select a max width for images in the ZIP file (512px, 768px, or 1024px)
- Extract all files to a single folder for training, then head to kohya_ss to train or fine-tune your Stable Diffusion model
Uploading an image:
Uploading a video:
Downloading a ZIP file:
MIT License
If you have any questions or find any issues, please create an Issue with detailed information.
- Please note this has only been tested on a Windows computer using Firefox. Your mileage may vary on other OSs and browsers.
Anyone is welcome to add new code or features! Please create an Issue first, comment that you're working on it (so multiple people don't work on the same thing), and mention the Issue number in your Pull Request.