How can I enforce vertical composition for 1024x1792 images in DALL-E 3?

Below is some guidance on this matter:

• Model Behavior & Parameters
  DALL·E 3 is designed to generate images with the dimensions you request (e.g. 1024x1792 for a vertical format). However, the underlying model decides on composition based on learned patterns, meaning that simply setting the resolution doesn't strictly enforce the “vertical” composition.

• Prompt Engineering
  While you can try adding cues like “portrait format” or “vertical composition” to your prompt, there isn’t an officially supported or reliable method to force the model to produce a strictly vertical image. The model’s visual output is influenced by its training and, in some cases, might default to a horizontal layout—even with appropriate aspect ratio settings.

• Inherent Behavior
  This may be viewed more as an inherent characteristic of the model’s training and image composition process rather than a bug. The API allows you to set the image dimensions (as documented), but finer control over image orientation isn’t explicitly provided.

• Next Steps
  If you find this behavior problematic for your application, you might consider:

  • Experimenting with different descriptive modifiers in the prompt, though results may vary.
  • Adjusting post-generative crop or layout adjustments in your app.
  • Providing feedback to OpenAI if you believe this aspect of the model could be improved with more explicit compositional control.

For now, there’s no dedicated parameter or guaranteed prompt technique to ensure a strict vertical image composition with a 1024x1792 size.