Yup. You still have to do some manual work if you want to create something like a comic and make sure each panel is internally consistent with the world. Heck, even OpenAI’s own cookbook demonstration of the dog comic was a complete mess. I ended up having to make a lot of manual edits just to fix it.

I also think most people fall into the Fiverr camp: they need something that’s “good enough,” and don't have a particularly critical eye for detail - in which case even one-shotting with the latest models, gpt-image-2 or nb2 can get you there.