Multimodal Fine-Tuning Data Generator

Upload images to generate JSONL training data for multimodal fine‑tuning

中文 | English

Overview

This project generates JSONL training data from images, with a simple frontend UI.

Demo

output.compress-video-online.com.2.mp4

Quick Start

One‑click scripts

Linux/macOS:

cd backend
bash start.sh

Windows:

cd backend
start.bat

FAQ

• Q: Which image formats are supported?
  • A: JPG, PNG, WebP. Recommended size < 10MB per image.
• Q: How to process a large batch of images?
  • A: You can upload multiple images at once; recommend ≤ 50 per batch.
• Q: Where are the generated files stored?
  • A: In backend/outputs/ on the server, and they are auto‑downloaded to your computer.
• Q: Can I customize the output data format?
  • A: Yes. Modify create_training_data in backend/app.py.

Contributing

Issues and PRs are welcome for feature improvements, bug fixes, and documentation.

Community

Explore our community (Chinese): 👉 Tech Community | Fufan Space