The problem with orientation has unfortunately been known for a long time and is still not fixed. DALL·E tries to place the subject as filling the image as much as possible, and then orients it in portrait mode. The opposite never happens. I’m not even sure if the developers themselves know why. I’ve created many images, and it keeps happening to me as well.
The following tricks might help.
Place an object next to the main figure that requires a lot of vertical space. In the last example, there is a streetlight. It probably helped.
Place a tall object in the description.
Also, describing an image from top to bottom often helps.
Describe the figure standing upright.
You many time have to trick DALL·E almost in to to do the right thing.
It is important to never use negations. DALL·E does not understand negations, everything that is mentioned is implemented in the image, and words like “not” or “without” do not work.
It is also important to know that what is written as a prompt is not necessarily what DALL·E has received. ChatGPT often modifies the prompt before sending it to DALL·E. It often makes sense to ask ChatGPT to display the prompt that was sent to DALL·E, to see exactly what DALL·E received, and then post this prompt in the forum.
(I cannot create images in the last days, don’t know why, but would otherwise try the prompts out.)
Portrait orientation it is still a bug in the DALL·E system. There are still several, you can find some tips and known bugs and work a rounds here.