I pay for both MJ and DALL-E (though OpenAI mostly gets my money for GPT) and don't find them to produce significantly better images than popular checkpoints on CivitAI. What I do find is that they are significantly easier to work with. (Actually, my experience with hundreds of DALL-E generations is that it's actually quite poor in quality. I'm in several IRC channels where it's the image generator of choice for some IRC bots, and I'm never particularly impressed with the visual quality.)
For MJ in particular, knowing that they at least used to use Stable Diffusion under the hood, it would not surprise me if the majority of the secret sauce is actually a middle layer that processes the prompt and converts it to one that is better for working with SD. Prompting SD to get output at the MJ quality level takes significantly more tokens, lots of refinement, heavy tweaking of negative prompting, etc. Also a stack of embeddings and LoRAs, though I would place those more in the category of finetuning like you had mentioned.
For MJ in particular, knowing that they at least used to use Stable Diffusion under the hood, it would not surprise me if the majority of the secret sauce is actually a middle layer that processes the prompt and converts it to one that is better for working with SD. Prompting SD to get output at the MJ quality level takes significantly more tokens, lots of refinement, heavy tweaking of negative prompting, etc. Also a stack of embeddings and LoRAs, though I would place those more in the category of finetuning like you had mentioned.