Also I think it really needs to let you actually try it with your own image (watermarked or whatever). Who wants to pay 5 pounds to find out the generated image doesn't quite look like you, or looks too fake or whatever.
The pricing seems maybe ok if you know you're definitely getting a usable image out of the process. But if there's any risk whatsoever, it feels much too expensive.
The before/after comparison is an obvious gap - I didn't include any example output on the page at all, which makes it near impossible to evaluate. That's being fixed today.
The free preview / watermarked trial is a good idea I hadn't fully thought through. The hesitation was cost (each generation costs real money to run) but a low-res watermarked preview before payment is a reasonable middle ground. Worth testing.
Thanks for taking the time to look.
The plan would be: generate once upfront, deliver a low-res watermarked version immediately, then serve the full-res after payment. One generation, two delivery tiers. The cost is the same either way - we're just changing when the user sees the output.