Multimodal image generation (Ethan Mollick)
๐
I can give GPT-4o two photographs and the prompt โCan you swap out the coffee table in the image with the blue couch for the one in the white couch?โ (Note how the new glass tabletop shows parts of the image that werenโt there in the original. On the other hand, the table that was swapped is not exactly the same). I then asked, โCan you make the carpet less faded?โ Again, there are several details that are not perfect, but this sort of image editing in plain English was impossible before.
Or I can create an instant website mockup, ad concepts, and pitch deck for my terrific startup idea where a drone delivers guacamole to you on demand (pretty sure it is going to be a hit). You can see this is not yet a substitute for the insights of a human designer, but it is still a very useful first prototype.
It is impressive.