9 Better Image Generation with Flux

 

This chapter covers

  • Introducing the Flux diffusion model.
  • How to evaluate new models.
  • Understanding Prompt Adherence.
  • Limitations and strengths of different models.
  • Considering the right model for a task.
“Coofffeee!”

-- Agent Dale Cooper (as “Dougie Jones”), Twin Peaks: The Return.

In this chapter we’ll be introducing a very exciting development in the Stable Diffusion world: Flux. Flux is a very powerful diffusion model developed by Black Forest Labs, a company founded by many of the original members of Stability.ai, the company behind Stable Diffusion. Our chapter’s opening quote is from Twin Peaks: The Return, a third season of the show Twin Peaks (referenced in our first chapter) that appeared 25 years after the end of the original. Flux is to Stable Diffusion much like Twin Peaks: The Return is to the original series: fundamentally the same, but also something entirely new and different.

9.1 Why Flux

9.1.1 Enhanced Image Quality

9.1.2 Better Human Anatomy

9.1.3 High Fidelity Text Generation

9.1.4 Prompt Adherence

9.2 Why Stable Diffusion 1.5 and XL?

9.2.1 Resource Requirements

9.2.2 Style

9.3 Conclusion

9.4 Summary