Suggestion: Official Turbo model
First off great model! The reason I'm suggesting a Turbo model is simply because on mid-range/lower-end hardware this thing is super sluggish.
3060ti based rig - INT8 version - 720p/50 steps - 52 seconds
5090 based rig - INT8 version - 1080p/50 steps - 12 seconds
Of course if running the non quantized version the above would be at least 2X slower. As good as the model is the popularity will be limited as the majority of users fall into the low-end/mid-range category. So even if they love it they're not going to stick around or bother with it unless they are very patient people. There is and will be speed lora's, however more often than not they are subpar and/or very slop-ish...an official Turbo model that retains the original models vision win every time.
Or:
-The Base model as it currently is with all it's soul at 50 steps and cfg etc.
-The SFT released as an official 8-step Turbo model with the output looking as it does now.
If an official Turbo version happens, side-by-side text rendering and complex layout tests would tell us more than timings alone.