Portals TM kind of represents a specific interpretation of how a set of portals would be used. If you didn't, I'd try emphasizing that it's identical twins wearing the same clothes.
The model probably hasnt seen this alot, so if you can modify some images to look like what you want you can train like StableDiffusion XL base model on them and get better results with like 10-20 images.
But as this is DallE SaaS, not alot you can do.
Maybe try giving the model such a modified image and tell it to refine it with img2img.
This is why I think AI is a good thing for artists because this is a good start. It wouldn't take much for an artist to alter it now to get what you want. Saved a lot of time which is the point people lose sight of.
Maybe try adding the Portal 2 name into the prompt. I don't use DALL-E as the FOSS solutions just give me so much control, but I suspect that the names Portal/Portal 2 and Aperture Science are in the corpus for DALL-E. I will have to test my SDXL setup to see what it comes out with. Are you able to use Controlnets with DALL-E? If so, it might be worth doing a basic sketch and having it fill in details.