Composition planning
Choosing an AI Image Aspect Ratio
Aspect ratio changes what the model can fit into the frame. Choosing the destination first helps you protect the subject, reserve copy space, and avoid unnecessary crops later.
Choose the destination before the ratio
Flux2.Cloud exposes common ratios including 1:1, 16:9, 9:16, 3:2, 2:3, 4:3, and 3:4. Image-to-image workflows can also offer Auto, which follows the first reference when the selected model supports it.
The same prompt can produce a different composition when the frame changes. A vertical frame may show more of a person; a wide frame may expand the environment; a square frame often pushes the main subject toward the center.
Review checklist
- 1:1: square cards, profile concepts, product tiles, and balanced single subjects.
- 16:9: website heroes, presentation covers, video thumbnails, and wide environments.
- 9:16: stories, reels covers, phone-first posters, and full-height characters.
- 3:2 or 4:3: editorial photography, campaign scenes, and flexible landscape layouts.
- 2:3 or 3:4: portrait posters, book-cover concepts, and vertical product compositions.
Write composition instructions for the frame
Do not rely on the ratio alone. Tell the model where the subject should sit, how much of it should be visible, and where empty space is needed. “Full body, subject on the right, clean copy space on the left” is more useful than only selecting 16:9.
Photorealistic editorial portrait in soft morning light, waist-up subject positioned in the right third, looking toward camera, calm neutral interior, large clean negative space on the left for a headline, natural skin and fabric texture, 16:9 landscape.
Review the safe crop before publishing
Interfaces and social platforms may crop previews even when the original file is preserved. Keep important faces, products, and text away from the extreme edges, and preview the image in the real destination before publishing.
If one asset must serve several channels, generate separate compositions rather than stretching or blindly cropping a single image. The prompt can stay similar while the ratio and copy-space instruction change.
Apply the method to a real generation
Save the source and prompt, then change one variable per attempt. Before publication, inspect identity, hands, text, logos, and rights-sensitive material at full size.
Flux 2 Cloud