I haven't found SDXL to be inherently much slower than 1.5, besides the obvious 4x slowdown from having twice the linear resolution.

Yeah, in my experience it's actually FASTER because I was already doing high res gens, and that required 2 or 3 passes previously.

Well, its 2.1 instead of 1.5, but:

https://huggingface.co/stabilityai/sd-turbo

I assume that this was ready for release because Stable Video Diffusion (which is also an SD2.1-based model) is essentially this plus a motion model.)

I wouldn't be surprised if their hosted-only SD1.6 beta is, or has, a turbo version, and if that gets released publicly, that's where we'll see an SD1.x turbo.

It _would_ be nice if they offered 512/768/1024 px variants of the models. I frequently don’t actually need the full 1024 px as it just needs to look good enough for a chat thumbnail. Then I could upscale it manually later. There’s other models like Kandinsky but it’s not super convenient to use multiple models with different code and what not.

SDXL Turbo also uses a distilled version of SDXL so it gets a speed bonus from that too.

is it known how much larger SDXL is compared to SD1 or SD2?

SD1.x is around 1.1B parameters including VAE, SD2.x is slightly more (uses the same UNet and VAE, but a bigger text encoder; not finding stats as quickly as I'd like), and SDXL is 3.5B parameters (single model only, but StabilityAI's preferred base + refiner model setup is effectively 6.6B parameters -- some bits are shared.)