Image Output Specifications and Ratios
Set the destination before writing the prompt. Social portraits, ecommerce squares and landscape covers require different subject scale, whitespace and text space.
#Keep model specifications separate
The Chinese source records these model-specific size sets:
| Model family | Source-listed sizes |
|---|---|
| GPT Image | 1024x1024, 1536x1024, 1024x1536, auto |
| DALL·E 3 | 1024x1024, 1792x1024, 1024x1792 |
| DALL·E 2 | 256x256, 512x512, 1024x1024 |
These describe the source's model versions, not a promise that each legacy endpoint remains available. Do not pass one model's options to another.
#Plan the final crop
Use square images for products and thumbnails, portrait 4:5-like frames for feeds, landscape 16:9-like frames for covers/banners, and tall 9:16-like frames for stories and video covers. Not every final ratio is a native model size, so allow crop space.
#Match format to the endpoint
The source discusses PNG, JPEG and WebP, compression and transparency. Verify these options for the chosen model; transparency requires a format that supports alpha, such as PNG or WebP.
Retain a high-quality master for further editing. Produce lighter distribution files when the destination calls for them. Choose dimensions from the layout first rather than generating an attractive image that cannot fit it.
