10 The Flux Workflow
This chapter covers
- Understanding how to generate images with Flux in ComfyUI
- Using quantized models to save on memory
- Getting Flux to use styles with Prompt Guidance
- Diving into the details of Flux’s sampler
- Exploring other areas where Flux’s workflow differs from SD1.5 and SDXL
“But everything with you is seeing isn’t it?
Your primary sensory intake that makes your dream live is seeing”
--Will Graham, Manhunter (1986)
In the last chapter we learned that Flux is a powerful new open image generation built by many members of the original team behind Stable Diffusion. In this chapter we’ll take a deep dive into the Flux workflow, how it differs from the standard Stable Diffusion workflow, and various ways we can use Flux’s unique configuration options to get the most out of the base version of Flux. We’ll be using the currently most powerful open version of Flux, Flux.1-dev.
10.1 Getting Started
We will continue to use ComfyUI as our interface for working with Flux. Before we can generate an image on our own we’ll need to grab a copy of the Flux Workflow as well as several files needed to run Flux locally. Flux’s power comes at the cost of being very demanding of RAM so we’ll also be exploring ways that we can reduce these requirements while minimally sacrificing the quality of our results. This next section will walk through all of these necessary steps.