This is the result of merging several of my checkpoints. The faces are a bit more cartoony. Perhaps it's because of the addition of Noobai. The text comprehension is even better. DMD2 has been incorporated to accelerate image generation, saving you some computing power. With 6 steps, you can get an image. Adding LoRa takes 10-16 steps, and there's barely any improvement in quality above 20 steps.
My go-to settings: 20 steps, CFG 2.5, dpmpp_sde sampler, beat scheduler, 832x1536 resolution.
Steps: 6-30, CFG scale: 1-7, Sampler: dpmpp_sde/Euler a/dpmpp2m_sde,
Scheduler: beat/karras/sgm_uniform/exponential
Have fun!