Aura_Flow_0-3-Q4_0-4Bit

Accuracy vs. Size Tradeoff

Every variant is benchmarked against the FP16 baseline. Choose the right balance of accuracy and footprint for your workflow.

Usage Code Snippets

Which variant for your hardware?

Choose the right quantization path for your device and performance target.

About This Model

AuraFlow v0.3 is a fully open-source text-to-image foundation model developed by fal.ai. Built using a flow-based generative architecture inspired by Diffusion Transformers, AuraFlow focuses on semantic understanding, prompt fidelity, and compositional image generation. The model contains approximately 6.8 billion parameters and was designed as one of the largest Apache 2.0 licensed image generation models available. AuraFlow v0.3 supports multiple aspect ratios, high-resolution image generation up to 1536x1536 pixels, and achieves state-of-the-art performance on the GenEval benchmark for prompt following. The GGUF quantized version significantly reduces memory requirements while maintaining image quality, enabling efficient deployment in ComfyUI-GGUF and local inference environments. AuraFlow is particularly effective for concept art, illustrations, marketing visuals, creative prototyping, and complex multi-object scene generation.