Stable character generation pipeline progress update, need feedback

My custom diffusers pipeline is supposed to replicate automatic1111 but isn't quite right, can anyone look at the images and tell me whats off about them? It's 1216*832 generated and then upscaled with esrgan 2x. I think the Vae is being called properly, and dpm++ 2m karras, but these look less sharp than generating with automatic1111. Also my adetailer is a fork which is just inpainting the face after upscale… it looks off many times too compared to auto1111. *These are fixed seed and prompt constrained stable character generations with sarah peterson loras for the pose/scenes.