We’ve made optimizations and fine-tuned the existing Flux.1 DEV and Schnell versions with FP8 integration. These adjustments offer better support for text generation and content descriptions, allowing image generation with fewer steps. You can generate images with a minimum of 10 steps, though we recommend 15-20 or more to meet various user needs. This effectively reduces point consumption while slightly improving image quality. For best results, I suggest using it together with my original prompt generation tool, FLUX Prompt Pro, available for free in the ChatGPT GPTs store by searching for FLUX Prompt Pro.
Recommended Settings:
Steps: 10-30
CFG Scale: 3.5
Sampler: Euler
Scheduler: Simple
Clip Skip: 1
VAE: None
