The documentation on image generation with the new gpt-image-1 currently states that the mask leaves transparent areas ‘untouched’. This is factually incorrect, as @OpenAI_Support has stated in other topics complaining of the entire image being re-generated that this is a known limitation of the model. I recommend updating the documentation to something more accurate, such as ‘the model will focus changes on the area of the mask’, and also adding a note in the limitations section.
link to documentation
2 Likes