Lumina-Image 2.0
An open-source image generation framework for local use, with ComfyUI, Diffusers and fine-tuning support. Code uses the Apache 2.0 license.
Lumina-Image 2.0 is a local AI image generation framework for developers, researchers and people who want to generate images from text on their own hardware. It provides downloadable checkpoints, generation code and tools for adapting the model to your own image collections. The code uses the Apache 2.0 license.
ComfyUI support makes it an option for visual image workflows, while the Diffusers integration suits Python applications. A Gradio interface provides browser-based access to a locally running model. Batch generation handles collections of prompts, so you can produce images without submitting each request individually. An official Hugging Face Space also offers a hosted demo; that demo runs remotely rather than on your machine.
Fine-tuning support includes training on image and text pairs and LoRA adaptation through Diffusers. The companion Lumina-Accessory project builds on the model for controllable generation, image editing and identity preservation, with support for training on one task or several tasks together.
The model uses Gemma-2-2B to encode text and FLUX-VAE-16CH for image encoding and decoding. Access to Gemma requires Hugging Face authentication. Local generation can use downloaded checkpoints, including .pth weights. The Diffusers integration supports CPU offloading to reduce GPU memory use.
