large models are actually really expensive to run: either run it yourself, or put up with the fact that someone’s not going to pay a lot for people to just use because reasons?

You can get results as good or better than Midjourney with stable diffusion…but it’s not as streamlined as Midjourney is. you have to find the right model on Civitai.com, you have to craft the prompt right, you can use controlnet. And then if you want to go the extra mile, refine it with img2img.

Stable diffusion + Automatic1111 + custom models and LoRAs. Completely free, only takes an hour or so to set up, and is top of the line.

Not sure how you mean. Stable Diffusion is the golden standard right now and it’s free. You just have to run it locally on your own hardware for the full configurability and features. It’s a pretty steep learning curve, and requires a decent, discreet GPU.