Flux.1 Lite - A lightweight AI model from Freepik

What is Flux.1 Lite?

Flux.1 Lite is a lightweight AI model released by the Freepik team, currently in alpha. Based on an 8 billion parameter transformer architecture, Flux.1 Lite is derived from the FLUX.1-dev model. The model weight file size of Flux.1 Lite is 16.3 GB, reducing RAM usage by 7 GB compared to the original model, while improving running speed by 23% and maintaining the same accuracy (bfloat16). The development team aims to further optimize the model, running it on a 24 GB consumer-grade GPU while maintaining its original accuracy and running faster, making high-quality AI models accessible to everyone. The release of Flux.1 Lite allows users to experience high-quality image generation effects in resource-constrained environments.

Main features of Flux.1 Lite

• Image generationGenerate detailed images based on text descriptions.
• Resource optimizationReduces GPU memory usage, making it suitable for running on consumer-grade GPUs.
• Speed increaseCompared to the original model, the running speed is improved by 23%.
• Precision maintenanceIn the lightweight version, the same bfloat16 precision as the original model is maintained.
• Easy to deployIt runs on common hardware, making it easy for developers and users to deploy and use.

The technical principles of Flux.1 Lite

• Parameter optimizationBased on the individual blocks in the analysis model, remove or optimize the parts that contribute less to the final result to reduce the model size.
• Mean Square Error (MSE) AnalysisThe contribution of each block in the MSE quantization model to the output image is used to identify blocks that can be removed without affecting image quality.
• transformer architecture: Based on the transformer architecture for image processing and generation, a very successful model architecture in the field of natural language processing.
• Precision maintenanceDuring model compression, technical means are used to maintain the model's accuracy and ensure the quality of the generated images.
• Video memory and speed optimizationBased on optimized model structure and algorithms, it reduces memory usage, improves processing speed, and adapts to resource-constrained environments.

How to use Flux.1 Lite

Using Flux.1 Lite in ComfyUI is no different from the native Flux-Dev; you only need to change the Unet loading method. Download the model flx.1-lite-8B-alpha.safetensors and place it in the /ComfyUI/models/unet directory.

The HuggingFace model library of flux.1-lite-8B-alpha.safetensors:https://huggingface.co/Freepik/flux.1-lite-8B-alpha/blob/main/flux.1-lite-8B-alpha.safetensors

Application scenarios of Flux.1 Lite

• Personal entertainment and creationUsers can use Flux.1 Lite to generate personalized images, such as anime-style characters and landscapes, for personal entertainment or social media sharing.
• Education and LearningIn the field of education, Flux.1 Lite serves as a teaching tool to help students understand the principles of image generation and the application of AI technology.
• Game developmentGame developers can use Flux.1 Lite to quickly generate in-game image resources such as characters and scenes, accelerating the game development process.
• Digital ArtArtists and designers use Flux.1 Lite to create digital artworks and explore new art styles and forms of expression.
• Advertising and MarketingMarketers use Flux.1 Lite to generate compelling advertising images for use on social media, websites, or print materials.