I wish this is something Ai services would understand. You don't charge bases in your teach stack. You charge based on your use case.

Tattoos are not something that you get every day so it makes no sense to pretend that your customers will use it like chat gpt. Costing more makes sense since it's a one off transaction that needs to cover overheads.

There is a whole raft of services which would seriously benefit from doing the same.

Undermind is one I can't justify buying because I do two literature reviews a month at most and have no use for an ongoing subscription I'll forget about. But I'd be happy to pay $5 per search.

I’m curious why you don’t think tattoo shops would want this as a monthly expense? $30/mo is cheap to give customers the ability to easily generate designs. Like another commenter said in this thread, good artists are often booked months out and that’s just to get the design started. Make a simple app that shows the UI in kiosk mode on an iPad using the store’s account. Even offer a finetuning service that they can use with photos of their own designs to get a unique style per customer (I’ve successfully finetuned myself with under 20 photographs, although style might be better with 50-100).

The other big benefit of tattoo shops is that they’re a well established type of business that you can sell to, instead of trying to market to consumers. First thing I’d do is buy a list of tattoo parlors and their contact information from InfoGroup/Data Axle/Dun & Bradstreet and start cold contacting them with a free month so they can try it out. I’d monitor the demos and reach out to those who don’t use their free month much to get feedback and refresh the demo period so they can try again. There are conventions where you can sell to the owners and artists en masse too. Google says there are over 20,000 stores in the US which is enough scale to just have a few GPUs to handle base load and autoscale during peak to drive costs down.

I think it can be a very lucrative lifestyle business to someone with good distribution and sales sense, just as long as some VC doesn’t come in and prop up a better resourced competitor.

> for people that don't want to set up a local model for whatever reason, it seems reasonable.

The reason being that in order to get a GPU that doesn't commit sudoku the moment you install SD, you need to shell out way more than the price of a single tattoo.

That's why people who own many 3d printers make money. If all it takes to print a model is one 3d printer, why doesn't literally every single person on the planet own one?

For sure, cost is maybe the most common reason.

But I have friends who game on cards that can run SD no problem (prior to a recent upgrade, I had SD running on a 2060S with no problem, which is like $250?), they just don't want to for other reasons (laziness, hard drive space, feel that it is too complicated, etc.).

> commit sudoku

I hate it when my GPU decides to waste time on puzzles instead of rendering things.

I like the idea of locking in the seed and params. Will see what I can do