lycoris-lora

What it is and what it does

LyCORIS is a PyTorch library that wraps Stable Diffusion models with parameter-efficient fine-tuning adapters. Instead of training all model weights, it applies low-rank or other structured modifications to a subset of layers, dramatically reducing memory and storage requirements while maintaining output quality. The library supports multiple adaptation algorithms (LoRA, LoHa, LoKr, DyLoRA, and native fine-tuning) with different trade-offs between fidelity, flexibility, diversity, and model size.

You can use it three ways: as a training backend for kohya-ss scripts (the most common path for Stable Diffusion practitioners), as a standalone PyTorch wrapper for custom training loops, or as pre-trained adapter weights loaded into inference tools like sd-webui, ComfyUI, or InvokeAI. The library handles the adapter lifecycle—creation, application to models, stacking multiple adapters, and restoration—leaving your base model untouched.

Use it for:

• Train custom Stable Diffusion adapters on specific visual styles or concepts using kohya scripts without full model fine-tuning.
• Reduce storage footprint of fine-tuned models from gigabytes to megabytes by using low-rank adapters instead of full checkpoints.
• Stack multiple LyCORIS adapters on a single base model to combine different trained concepts or styles at inference time.
• Integrate parameter-efficient tuning into custom PyTorch training pipelines via the standalone wrapper API.
• Fine-tune Stable Diffusion models on consumer GPUs with limited VRAM by using adapter-based training instead of full model training.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Implements parameter-efficient fine-tuning algorithms (LoRA, LoHa, LoKr, DyLoRA, and others) for Stable Diffusion models, enabling training and inference with reduced memory and storage overhead.

Yes. LyCORIS is actively maintained, has no known vulnerabilities, installs cleanly with minimal dependencies, and is widely adopted in the Stable Diffusion community. Choose it if you need to train or use parameter-efficient adapters for Stable Diffusion; the multiple algorithm options and integration with popular training frameworks make it the standard tool for this task.

Install

lycoris-lora on PyPI

pip install lycoris-lora

uv add lycoris-lora

poetry add lycoris-lora

Installing lycoris-lora

Before you install

Low friction: pure Python wheel with four straightforward runtime dependencies (torch, einops, toml, tqdm). Actively maintained with recent commits and a stable release cadence.

License in practice

Apache-2.0 permissive license allows commercial and private use with minimal restrictions, making it suitable for both open-source and proprietary projects.

Quickstart

pip install lycoris-lora

from lycoris import create_lycoris, LycorisNetwork

LycorisNetwork.apply_preset({"target_name": [".*attn.*"]})
lycoris_net = create_lycoris(
    your_model, 1.0, linear_dim=16, linear_alpha=2.0, algo="lokr"
)
lycoris_net.apply_to()
output = your_model(x)

Requires PyTorch and a compatible GPU (CUDA 11.8, 12, or 12.1 recommended). Python >= 3.10 required.

Verify before relying

• Whether all newer LyCORIS model types are supported across all listed inference platforms (ComfyUI, InvokeAI, CivitAI, Tensor.Art) or if some lag behind.
• Specific CUDA version compatibility constraints beyond the classifiers (11.8, 12, 12.1) listed in metadata.

Package facts

Evidence: lycoris_lora-3.4.0-py3-none-any.whl

Tags

More Artificial Intelligence packages

Further reading