How to choose an AI image size
- Pick the aspect ratio you need, or choose where the image will be used, such as an Instagram post, a story or a YouTube thumbnail.
- The table shows the closest supported size for popular image generators, with the exact parameter to use and how much you would need to crop if the model cannot match the ratio exactly.
- Copy the size into your prompt, API call or interface settings.
Why size matters for image generation
Image models are trained at particular resolutions, and most only accept certain sizes. OpenAI's GPT Image and DALL·E 3 offer three fixed sizes: square, landscape and portrait. Stable Diffusion XL works best at about one megapixel in specific "bucket" sizes such as 1216×832, and SD 1.5 around 512 pixels; going far outside these produces duplicated subjects, stretched bodies or extra limbs. FLUX accepts most sizes in multiples of 16, Midjourney takes any ratio with --ar, and Google's Imagen and Ideogram offer a list of preset ratios.
Generating at the right ratio is better than cropping afterwards, because composition is planned for the frame: a 9:16 prompt puts the subject in a tall frame instead of cutting it off. For print or large displays, generate at the native size and upscale with a dedicated upscaler. Supported sizes change as models are updated, so check the provider's documentation for the latest options. This page is a reference and calculator; nothing is generated or sent anywhere.
What size should I use for Instagram?
1:1 or 4:5 for feed posts and 9:16 for stories and Reels. Generate at the closest supported size and crop lightly.
Can DALL·E 3 make 16:9 images?
Its closest size is 1792×1024, which is 7:4. Cropping about 2% gives an exact 16:9 image.
Why do my SDXL images look strange at large sizes?
SDXL was trained around one megapixel. Larger sizes often repeat subjects; generate at a bucket size and upscale instead.