Short answer: set the aspect ratio before you generate, for the place the image will go: 4:5 for an Instagram or LinkedIn feed post, 9:16 for Stories, Reels, TikTok and Shorts, 16:9 for YouTube, slides and website headers, 2:3 for Pinterest, 1:1 for profile pictures and 1.91:1 for link previews. Cropping a finished image to a new shape cuts into the composition the model planned.
Which ratio goes where?
| Where it goes | Aspect ratio | Typical size in pixels |
|---|---|---|
| Instagram feed post | 4:5 (or 1:1) | 1080 × 1350 |
| Instagram Stories, Reels, TikTok, YouTube Shorts | 9:16 | 1080 × 1920 |
| YouTube video and thumbnail | 16:9 | 1280 × 720 or larger |
| Slides (PowerPoint, Google Slides, Keynote) | 16:9 | 1920 × 1080 |
| LinkedIn feed post | 4:5 or 1:1 | 1080 × 1350 |
| Link preview (Open Graph, LinkedIn, Slack) | 1.91:1 | 1200 × 630 |
| Pinterest pin | 2:3 | 1000 × 1500 |
| X (Twitter) post | 16:9 | 1600 × 900 |
| Profile picture, avatar | 1:1 | 400 × 400 or larger |
| Blog featured image | 16:9 or 3:2 | 1600 × 900 |
| Website hero banner | 16:9 or 21:9 | 1920 × 1080 or wider |
| Print, A4 or A3 | about 5:7 | depends on print size |
Platforms change their recommended sizes from time to time, so check the help page of the one you post to before a big campaign. The ratios above have been stable for years.
Why set the ratio before you generate?
A model plans the whole picture for the shape you ask for. In 9:16 it stacks the subject and the space above and below it. In 16:9 it spreads the scene sideways and leaves room for a headline. Crop a square image to 9:16 and you lose two thirds of the width; crop a 16:9 image to 4:5 and the subject may not fit.
So set the ratio first, and make one version per shape when the same image goes to several places. A campaign often needs three: 4:5 for the feed, 9:16 for Stories and 16:9 for the website.
How much room should you leave for the interface?
Some places cover part of your image:
- Stories and Reels put the profile name at the top and buttons and captions at the bottom. Keep faces and text out of the top and bottom 250 pixels or so, roughly the outer 14%.
- Profile pictures are shown as a circle. Centre the face and leave margin around the head.
- Instagram's profile grid shows feed posts cropped to 3:4, so keep the important part inside the middle of a 4:5 image.
- YouTube thumbnails get a timestamp in the bottom right corner.
- Website heroes are cropped narrower on phones. Keep the subject near the centre, or make a separate mobile image.
Write this into the prompt: "subject in the centre, the top and bottom fifth kept clear of faces and text".
How do you set the ratio in each model?
| Model | How to set it | Notes |
|---|---|---|
| Midjourney | --ar 4:5 at the end of the prompt | Any whole-number ratio works, such as --ar 5:7 |
| Flux | Width and height, or a ratio menu in the tool | Sizes are usually rounded to multiples of 16 or 32 |
| Nano Banana (Gemini) | A ratio setting where the tool has one, or say it in the prompt | When editing, it tends to keep the shape of the photo you attached |
| GPT Image | Square, landscape or portrait | Only 1:1, 3:2 and 2:3; crop to other shapes |
| Stable Diffusion, SDXL | Width and height | Generate near 1 megapixel (such as 1344 × 768), then upscale |
| Ideogram | A ratio menu | Many ratios, including 9:16 and 16:9 |
Three tips that save generations:
- Nano Banana keeping the old shape? Crop your input photo to the target ratio before you attach it, then ask for the edit.
- GPT Image and 16:9? Ask for landscape (3:2) with "extra empty space at the top and bottom", then crop to 16:9 without cutting anything important. For 9:16, ask for portrait (2:3) with space at the sides.
- SDXL and full HD? Do not ask for 1920 × 1080 directly; the model was trained around one megapixel and repeats the subject at large sizes. Generate at 1344 × 768 and upscale.
Can you change the ratio afterwards?
If you love an image in the wrong shape, extend it instead of cropping. Midjourney has zoom out and pan, Photoshop has generative expand, and Nano Banana and GPT Image will "extend the image to 16:9, continuing the background naturally". Check the new edges: extended areas sometimes repeat patterns or bend straight lines.
Where can you check this?
- YouTube Help: Video resolution and aspect ratios
- Midjourney documentation
- Google AI for Developers: Image generation with Gemini
What is the short version?
- 4:5 for feeds, 9:16 for Stories and short video, 16:9 for YouTube, slides and the web, 2:3 for Pinterest, 1:1 for avatars, 1.91:1 for link previews.
- Set the ratio before you generate, and make one version per shape.
- Keep faces and text away from the edges that the interface covers.
- Use your model's own ratio setting; for GPT Image and SDXL, generate close and crop or upscale.
In the Prompt Gallery, the Customize panel puts the ratio you pick (1:1, 4:5, 3:2, 16:9 or 9:16) into the prompt you copy, replacing the one it had, Midjourney's --ar included. The course lesson on customizing an image prompt shows how, and Midjourney, Flux, Nano Banana or GPT Image helps you pick the model.
