Greetings Everyone, I hope your New Year 2024 is STELLAR!
This model is a Turbo variant, so the best results will be between 6 and 12 steps (step count) and with a config (cfg) between from one up through two (1-2).
This model is amazing at extending between dimensions, as in, this model can create great results from latents that are different resolution ratios than those you upscale to. So, an image that is 1024x1024 could be upscaled then sent to another image to image variety of sampling to resolutions like 832x1280 or even 1280x2024.
This is a Turbo model, so you must use it with the Turbo setting, or in ComfyUI, with the Turbo sampler node which will let you select steps and de-noising values then pass the sigmas to the ksampler.
Please enjoy this checkpoint, and once again, Happy New Year!