For the first one, I'd be making a hypernetwork/textual inversion like the bad-prompt one I use. I have little experience in it, and frankly, the learning process is not worth it to me. Making a LoRA for normal height is fine, but reducing LoRA usage as much as possible tends to produce better images (IMO). LoRA's are meant to narrow down prompting that can't be gotten across with normal prompt methods, so I'd try more varied and weighted prompts.

The second one seems possible. It is true that stable diffusion struggles with complex descriptions for more than one thing, but I think it'd turn out well enough. The dataset would have to be more curated than my usual gathering process, but I'll look into it.

I can do Orc Girls.

These are all great ideas... for someone who has more time than me.

The problem with training for exact poses (the kind you get from controlnet) without using controlnet, is that it's really really difficult. You can get a general pose, like the ones I provide, but unless you have really good and clean training data that doesn't deviate from the pose, it's not happening. Even then, you'd have to hand-pick your dataset if you want a specific pose that isn't a sex position / really iconic pose.

Victory and Defeat poses are definitely possible, I imagine they would be easier to gather a dataset for and it would lend well to training for a LoRA. It would be pretty general in use so nothing too specific.

An automatic sprite template could be taken a bunch of ways, all of which sound like they'd kinda fall flat. Getting the dataset for a LoRA of it would be difficult, and other ways don't sound that promising, other than controlnet in conjunction with a LoRA.

I could imagine doing a dating sim pose LoRA template, unless you mean the specially drawn artwork that pops up in a VN every now and then.

Incorporating AI into comic/manga art sounds like a lot of hassle, even if it were geared to do so with a model specifically trained for it.

already exists I think

@ChimpEnjoyer Yes, you are right. But I have a feeling that other Lora's are not of the best quality... from looking at the shape of the nipples, etc.