🧩 OneShot Dataset Prep – Minimal Input, Maximal Output

OneShot Dataset Prep is a lightweight, web-based tool that transforms a single image into a complete, structured dataset – ideal for one-shot LoRA training, quick concept sets, and testing workflows.

Upload your image → smart crop & flip augmentation → instant ZIP download with 14 labeled images.

All processing happens locally on your server –
no GPU required, no cloud tricks, no nonsense.

“From one image to one dataset – minimal input, maximal output.”

✅ Project Status

All core features are finished and production-ready.
The project is now fully public and open-source:

✔️ Core Image Processing (Crop & Flip Logic)
✔️ Backend API (Flask-based ZIP generation)
✔️ Frontend UI (Tailwind uploader with matrix-style aesthetics)
✔️ Teamspace Support (Archive separation, access control, quota management)

🤝 Teamspace – Built for Collaboration

OneShot Dataset Prep includes integrated team archives – designed for shared projects, dev groups, and collaborative dataset pooling:

✅ Create and manage teams (permission-based)
✅ Upload directly into team storage via dropdown
✅ Each team can store up to 50 datasets, separate from personal quotas
✅ Team members automatically share access to team archives
✅ Visual feedback when storage limits are reached (upload lockout & highlight)
✅ Dedicated Team Archive page for clean management

“No more sharing by hand – just upload once and the whole team’s got it.”

🧠 Best Practice: One-Shot That Works

Yes, 1 image = 14 crops.
But for better real-world LoRA results, use two angles:

👉 Recommended Setup:

1x front-facing view
1x back-facing or alternate angle

This gives you 30 images (2 originals + 28 variants), improving generalization and spatial consistency – especially for characters.

“Two views, one dataset – double the context, double the performance.”

🚀 Public Release – Get Started Now!

The testing phase is over –
the tool is now fully public and available on GitHub.

👉 GitHub Repo
Includes Docker support, maintainer scripts, and full documentation.

Feedback, bug reports, and ideas are welcome – just open an issue or drop a PR!

🧰 Key Features

Drag & drop upload interface
Automatic generation of 14 image variants (crop & flip)
Instant ZIP packaging – ready for training
Clean file naming (_top_left_flip, bottomhalf, etc.)
Multi-user system with login, admin panel & registration control
Quota system:
- Personal archive = up to 10 datasets
- Team archive = up to 50 datasets
Dynamic upload lockout when quota is reached
Matrix-style dark UI with Tailwind CSS 😎

📂 Example Output (from `sample.jpg`)

sample_original.jpg
sample_original_flip.jpg
sample_top_half.jpg
sample_top_half_flip.jpg
...
sample_bottom_right_flip.jpg

🛠️ Tech Stack

Component Technology Backend Flask 3.x Image Processing Pillow (PIL) Packaging zipfile + io.BytesIO Frontend HTML5 + Tailwind CSS Temp Handling tempfile Runtime Env Python 3.12