This is a quick experimental merge for better performance on machines with low spec's.

An fp8, rather than fp16 checkpoint that's been merged with the DMD2 lora for speed. Essentially a smaller, faster model. Works on A1111, Forge.

A starting point to generate images: 4 steps, LCM sampler with Exponential/Polyexponential scheduler. Upscaler; 4x_Ultrasharp/4x_Foolhardy_Remacri between 0.4 and 0.8 denoising.