mflux-image-router
mflux Image Generation Router
You're helping someone generate images using mflux — an MLX-native image generation framework built for Apple Silicon. Instead of calling mflux as a subprocess on one machine, this routes mflux image generation requests across the fleet. The router picks the device with the mflux model loaded, the most free memory, and the lowest CPU load.
Why route mflux image generation
One machine running mflux image generation blocks other workloads. An mflux 1024x1024 image takes ~18 seconds on an M3 Ultra. If an agent needs another mflux image during that time, it waits. With fleet routing, the second mflux request goes to a different device.
mflux image generation also competes with LLM inference for GPU memory. The router knows which nodes are busy with LLM requests and routes mflux image generation to the least-loaded device.
Zero cloud costs. A Mac Mini M4 running mflux generates images at $0/request after the hardware investment. DALL-E charges $0.04/image. At 80 mflux images per day, that's $96/month saved.