Lorata is a data labeling tool for the GenAI models. We plan to support the following labeling tasks:

• Text-to-Image
• Image-to-Image (Image editing, support multiple source images)
• Text-to-Video
• Image-to-Video
• Video-to-Video (Video editing)

Lorata has a built-in image/video editor that supports:

• Draw on image
• Crop image
• Trim video
• Generate image/video caption with AI
• Edit image with cloud AI models (e.g. Nano Banana)
• Extract image frames from video
• Export videos with different FPS

image-editing.mp4

video-trimming.mp4

Lorata supports exporting tasks as zip files, with the following structure:

# Text-to-Image task
task-name.zip
├── instructions
│   ├── item-1.txt
│   ├── item-2.txt
│   └── ...
├── images
│   ├── item-1.jpg
│   ├── item-2.png
│   └── ...

# Image-to-Image task (multiple source images)
task-name.zip
├── instructions
│   ├── item-1.txt
│   ├── item-2.txt
│   └── ...
├── sources_1
│   ├── item-1.jpg
│   ├── item-2.png
│   └── ...
├── sources_2
│   ├── item-1.jpg
│   ├── item-2.png
│   └── ...
├── targets
│   ├── item-1.jpg
│   ├── item-2.png
│   └── ...

# Text-to-Video task
task-name.zip
├── instructions
│   ├── item-1.txt
│   ├── item-2.txt
│   └── ...
├── videos
│   ├── item-1.mp4
│   ├── item-2.mp4
│   └── ...

# Image-to-Video task
task-name.zip
├── instructions
│   ├── item-1.txt
│   ├── item-2.txt
│   └── ...
├── sources
│   ├── item-1.jpg
│   ├── item-2.png
│   └── ...
├── targets
│   ├── item-1.mp4
│   ├── item-2.mp4
│   └── ...

Getting Started

Install pnpm and ffmpeg, then run:

cp .env.example .env
pnpm install
pnpm db:push
pnpm dev

(Optional) If you need image/video captioning model, configure .env with your OpenAI-compatible API endpoint:

OPENAI_API_MODEL= # Make sure to use a vision language model
OPENAI_API_KEY=
OPENAI_API_BASE_URL=

Training Frameworks

• ai-toolkit by @ostris
• sd-scripts by @kohya-ss
• musubi-tuner by @kohya-ss
• diffusion-pipe by @tdrussell