Image Generation

Generate hyper-realistic images from a text prompt, fully offline. A thin wrapper drives mflux (a native-MLX port of modern image models); you (the model) shape the brief into a strong prompt, generate a couple of candidates, let the user pick, and optionally upscale the winner for hi-res.

This is the programmatic, local equivalent of an app like Draw Things: local model weights, a scriptable Python API, no GUI and no cloud.

flowchart LR
  Brief["brief: subject, style, lighting, lens"] --> Gen["generate_image.py<br/>Z-Image-Turbo (MLX), 9 steps"]
  Gen --> Png["candidate png(s)"]
  Png -.->|optional| Up["upscale_image.py<br/>SeedVR2 super-resolution"]
  Up -.-> HiRes["hi-res png"]