Leaxer Model Registry

Central registry of AI models for Leaxer.

Overview

This repository hosts a JSON catalog of AI models with download URLs. The actual model weights are hosted on HuggingFace and other external sources.

Leaxer fetches this registry to populate its model browser/downloader.

Registry URL

https://raw.githubusercontent.com/leaxer-ai/leaxer-model-registry/main/registry.json

Categories

Category	Description
llms	Large Language Models (Llama, Qwen, Mistral, etc.)
checkpoints	Image generation models (SDXL, FLUX, SD 1.5)
loras	LoRA adapters for fine-tuning
vaes	Variational Autoencoders
controlnets	ControlNet models for guided generation
upscalers	Image upscaling models
text_encoders	CLIP, T5, and other text encoders
embeddings	Textual embeddings (negative prompts, etc.)

Model Schema

{
  "id": "unique-model-id",
  "name": "Human Readable Name",
  "description": "What this model does",
  "size_bytes": 1234567890,
  "size_human": "1.2 GB",
  "format": "gguf|safetensors|pth",
  "license": "apache-2.0|mit|openrail++|...",
  "commercial_use": true,
  "recommended": false,
  "tags": ["tag1", "tag2"],
  "download_url": "https://huggingface.co/...",
  "homepage": "https://huggingface.co/..."
}

Optional Fields

• base_model: For LoRAs/ControlNets - which base model they're compatible with
• quantization: For GGUF models - Q4_K_M, Q8_0, etc.
• context_length: For LLMs - max context window
• scale: For upscalers - 2x, 4x, etc.

Contributing

1. Fork this repository
2. Add your model to the appropriate category in registry.json
3. Ensure all required fields are filled
4. Submit a Pull Request

Guidelines

• Models must be publicly downloadable
• Include accurate file sizes
• Specify correct license information
• Use descriptive tags for discoverability

License

This registry is MIT licensed. Individual models have their own licenses as specified in the license field.