IMAGEKA makes images in three shapes: square, 4:3 landscape and 3:4 portrait. Social platforms ask for at least five. This guide shows which shape to generate for each common format, how to crop without losing the important part of the picture, and how to keep a feed looking consistent.
- Generate in the IMAGEKA shape closest to your target, then crop. 3:4 covers 4:5 and 2:3 posts, 4:3 covers 16:9 thumbnails, and 1:1 covers square posts.
- For extreme shapes (9:16 stories, 16:9 banners), write the prompt so the subject sits in the centre with empty space around it. That empty space is what you will crop away.
- Think in aspect ratios, not pixel sizes. Platforms change their recommended pixel sizes, so check their current help pages.
- Leave deliberate negative space for text, and add the text yourself in a design tool.
- Keep one anchor phrase, one palette and one preset for a recognisable feed. Label AI images where the platform asks, and write alt text.
The three shapes IMAGEKA makes
In Settings you can pick 512×512, 768×768 or 1024×1024 (square, 1:1), 1024×768 (landscape, 4:3) or 768×1024 (portrait, 3:4). There is no custom size, so every non-matching format means cropping, or adding canvas in an editor.
Two things follow from that:
- Always use the largest size of the shape you need. 512×512 is fine for quick tests, but cropping throws pixels away, so start with as many as you can. (Requests that go to one of IMAGEKA's specialist models are generated at 768×768 when you pick 512×512, so a test image may come back larger than expected.)
- Plan the crop in the prompt. The model does not know you will cut the sides off later. If you tell it "the subject is centred with empty background on both sides", the crop will not slice through anything important.
Which shape to generate for each format
These are the common aspect ratios behind most social formats. Exact pixel recommendations vary by platform and change from time to time, so treat the ratio as the fixed part and check the platform's own help page for current numbers.
| Target format | Typical use | Generate at | What you crop |
|---|---|---|---|
| 1:1 square | Feed posts on most networks, profile pictures | 1024×1024 | Nothing (profile pictures are often shown as a circle, so keep the face centred) |
| 4:5 portrait | Portrait feed posts, which take up more screen height than squares | 768×1024 | A thin strip, about 6% of the height, from top and/or bottom |
| 2:3 portrait | Pins and tall feed images | 768×1024 | About 11% of the width from the sides |
| 9:16 vertical | Stories, reels and short-video covers | 768×1024 | A quarter of the width from the sides, or extend top and bottom |
| 16:9 landscape | Video thumbnails, banners, link previews | 1024×768 | A quarter of the height from top and/or bottom |
| Very wide banners (around 3:1 or wider) | Profile headers and cover images | 1024×768 | Most of the height; better to extend the sides |
The maths: a 768×1024 image cropped to 4:5 keeps the full 768 px width and a 960 px height, so you lose only 64 px. Cropped to 9:16 it keeps the full 1024 px height and a 576 px width. A 1024×768 image cropped to 16:9 keeps the full width and a 576 px height.
Seeing the crop zones
Square and portrait feed posts
For square posts, generate at 1024×1024 and you are done. For 4:5 portrait posts, generate at 768×1024. You will trim only a sliver, so write the prompt as you normally would and leave a little headroom: "with space above her head" stops the crop from clipping hair.
Some profile grids show a cropped preview of each post, which can be squarer or taller than the post itself. Keep the subject near the centre so it still reads well in the grid.
Stories, reels and short-video covers (9:16)
9:16 is much taller than anything IMAGEKA makes. You have two options.
Option 1: generate 3:4 and crop the sides
From 768×1024, a 9:16 crop keeps the full height and the middle 576 px of the width. Describe a narrow, centred subject with plain background on both sides: "a single tall lighthouse in the centre of the frame, empty sea and sky on both sides". The result is 576×1024. Platforms will scale it up to fill the screen, and on a high-density phone it will look slightly soft. That is usually fine behind text, less so for fine detail.
Option 2: extend the top and bottom
Put the 768×1024 image on a taller 9:16 canvas in a design tool and fill the empty bands with a matching colour, a gradient or a blurred, enlarged copy of the image. This keeps the whole picture and gives you natural bands for a headline above and a call to action below. It works especially well with Minimalist or flat-colour styles, because the fill blends in.
In vertical formats, the platform's own interface (profile name, progress bar, caption, reply box and buttons) covers parts of the top, bottom and right edge. Keep faces, products and any text in the middle of the frame, and preview the post in the app before publishing.
Thumbnails, banners and link previews (16:9 and wider)
Generate at 1024×768 and crop a quarter of the height. Write "wide shot" and describe space above and below the subject, otherwise the model may fill the frame vertically and the crop will cut off a head or a tabletop.
For very wide profile headers, cropping a 4:3 image would leave a thin strip with little detail. Instead, generate a scene with a simple background (sky, a wall, a gradient, a textured surface), place it on a wide canvas, and extend the sides with a matching colour or a mirrored, blurred copy. Profile pictures often overlap one corner of a header, so keep that area empty.
Thumbnails and text overlays
Thumbnails are viewed small and quickly. What works at full size often turns to mush at thumbnail size.
- One subject, high contrast. A single clear shape against a simple background reads at any size. Ask for it directly: "one large subject against a plain dark background".
- Plan negative space. Decide where your headline goes (usually the left or top third) and describe that area as empty: "the left third is a smooth, out-of-focus wall". The composition guide shows layouts that leave room for text.
- Add the words yourself. IMAGEKA's general-purpose model can render a short word or two in quotes, but a design tool gives you sharp, correctly spelled, on-brand type you can edit later. Very few creators should rely on AI-generated lettering for a title.
- Check contrast behind the text. If the background is busy, add a semi-transparent dark band or a soft shadow behind the type in your editor.
- Test small. Shrink the finished thumbnail to a couple of centimetres wide on your screen. If you cannot tell what it is, simplify.
Keeping a consistent feed look
Because IMAGEKA picks a new random seed for every run and has no seed control, consistency has to come from the prompt. Three habits do most of the work.
1. Use an anchor phrase
Write one sentence that describes your look (medium, lighting, palette and mood) and paste it unchanged at the end of every prompt. Change only the subject.
2. Name a fixed palette
Two to four named colours ("sage green, oat and terracotta") hold a feed together better than any single style. Use the same names every time. "Green" and "sage" are not the same instruction.
3. Pick one preset and stick to it
Each style preset adds the same fixed text to every prompt, which is a free consistency tool. Choose one that suits your brand (the styles guide lists what each adds), or use Default and keep the style inside your anchor phrase. Do not do both, because the two descriptions will compete. The preset also matters for a second reason: IMAGEKA chooses between several models (photo, anime, painterly and a fast general model) based on the preset and your wording, and each has its own look. Keeping the preset and style words identical keeps every post on the same model. Words such as "poster", "logo" or anything in quotes switch to the general model, so a post that adds them can look out of place in the grid.
Also keep framing consistent. If most posts are centred close-ups, an occasional wide shot looks like it belongs to another account. Save the prompts that worked in a notes file so you can repeat them later.
Labelling AI images
Several major platforms now ask creators to disclose realistic AI-generated or AI-altered content, and some add an "AI" label automatically when they detect it. The rules differ by platform and change often, but the common pattern is that realistic images of people, places or events need a label, while obviously stylised illustration is treated more leniently. Check each platform's current policy, and when in doubt, label. A short "Image created with AI" in the caption costs nothing and protects your credibility.
For ads, paid partnerships and anything that makes a factual claim, disclosure matters even more. See using AI images commercially for the legal side.
Accessibility: write alt text
Most platforms let you add alt text to images, often under an "advanced" or "accessibility" option. Screen readers read it aloud to people who cannot see the image. Some platforms generate alt text automatically, but it is often generic.
- Describe what matters for the post in one or two sentences: "A mug of hot chocolate on a snowy windowsill beside a folded knitted blanket."
- Do not paste your generation prompt. Words such as "85mm lens, photorealistic" mean nothing to the listener.
- If the image contains text you added, include that text in the alt text.
- You do not need to start with "Image of". The screen reader already announces that it is an image.
A five-minute workflow
- Decide the target format first, and pick 1024×1024, 768×1024 or 1024×768 from the table above.
- Write the prompt with the crop in mind: centred subject, stated empty space, and your anchor phrase.
- Generate four images per run and download the best one immediately. Images are stored only in your browser.
- Crop or extend in a free editor, add text in the negative space, and check it at small size.
- Post with alt text and an AI label where needed, and save the prompt for next time.
New to prompting? Start with how to write prompts that work. If something in the image goes wrong, the troubleshooting guide covers the common fixes.
Frequently asked questions
Can IMAGEKA generate 9:16 or 16:9 images directly?
No. The available shapes are 1:1, 4:3 (1024×768) and 3:4 (768×1024). Generate the nearest shape with space around the subject, then crop or extend the canvas in an image editor.
Is 1024 px big enough for social media?
For feed posts, yes. Platforms compress and resize uploads anyway. Heavily cropped images, such as a 576 px-wide story, will look a little soft on high-resolution screens. IMAGEKA has no upscaler, so if sharpness matters, run the image through a separate upscaling tool before posting.
Why not just upload the image uncropped and let the platform crop it?
Automatic crops usually cut from the centre and can slice through faces or text. Cropping it yourself means you choose what stays.
How do I make every post look like it comes from the same account?
Reuse one anchor phrase describing light, palette and mood, keep the same style preset, and keep framing similar. Change only the subject sentence.
Do I have to say an image was made with AI?
It depends on the platform and on how realistic the image is. Many platforms require labels for realistic synthetic content, and labelling is good practice anyway. Check each platform's current rules.
Can I put my brand name in the image with the prompt?
You can try with short words in quotes, but spelling and lettering style are not reliable. For brand names, add the text or logo in a design tool afterwards.