Only when ChatGPT replaced DALL-E with GPT-4o, I first asked it to generate a prompt based on the uploaded original image:
Write a prompt that will instruct an AI image generator to produce a similar image.
Then it returned such prompt:
Create an image of a small, charming red robot with large round green eyes and pink headphones splashing through a puddle during rainfall. The robot has a shiny circular blue light on its chest and tiny hands. It’s walking with one foot splashing in water, creating splash effects. The background is a blurred green landscape with rain droplets falling. The scene has a dreamy, magical quality with a shallow depth of field. Lighting should highlight the water splashes and raindrops, giving the image a whimsical, cinematic feel. Digital art style with vibrant colors and soft focus.
that allowed the new model to finally generate a decent image:
Isn’t it just adorable?
BTW, a fun fact: when asked ChatGPT to generate a script for processing songs, as an example of a song title it proposed Paranoid Android by Radiohead. This means it’s their favorite song.