The Right Aspect Ratio and Resolution for AI Photos

There is a specific kind of disappointment that comes from generating a photo you love and then watching the platform ruin it. The top of your head gets cropped off. The image goes soft the moment it uploads. The composition that looked balanced in the preview turns lopsided in the feed.

None of that is the model's fault, and none of it is fixable after the fact. It is decided at generation time, by two settings most people leave on default: aspect ratio and resolution.

This is the boring half of AI photography, and it is also the half that separates images that survive the trip to a feed from images that do not.

What Gemini Actually Gives You

Nano Banana Pro supports ten aspect ratios — 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, and 21:9 — at three resolution tiers: 1K, 2K, and 4K.

The tiers scale with the ratio rather than being fixed pixel counts. A 1:1 image is 1024×1024 at 1K, 2048×2048 at 2K, and 4096×4096 at 4K. A 16:9 image is 1344×768 at 1K, 2688×1536 at 2K, and 5376×3072 at 4K. The pattern is that the longest dimension grows with the aspect ratio, so a wide image at a given tier has more total pixels than a square one.

Two practical notes before the platform table. First, availability of the higher tiers depends on your plan and on which surface you are generating from — several people have hit 4K being available in one place and not another on the same account. Second, generating at a higher tier costs more, and for most social use it is unnecessary. 2K is the sweet spot for nearly everything below.

Aspect Ratio: Pick It Before You Write the Prompt

The mistake almost everyone makes is generating a beautiful square image and then cropping it to vertical. That does not work, because the model composed the shot for the frame you asked for. Crop it later and you are cutting into a composition that was balanced for different edges — you lose headroom, or the subject drifts off center, or the environmental detail that made the shot work ends up outside the frame.

Choose the ratio first, and let the model compose for it. Then, if you need the same shot in two formats, generate it twice rather than cropping once.

4:5 — the default for almost everything. Instagram feed posts display at 4:5 (1080×1350), which is the tallest ratio the feed permits and therefore the most screen real estate a still image can occupy. It is also the right ratio for a LinkedIn headshot, because it gives you room above and below the head so the circular profile crop does not clip you. If you only ever generate one ratio, make it this one.

9:16 — Stories, Reels covers, TikTok, Shorts. Full-screen vertical. The trap here is that platform UI overlays eat the top and bottom of the frame — captions, usernames, action buttons, progress bars. Compose for a safe area roughly in the middle 70 percent of the height, and tell the model so: "composed with the subject centered in the middle 70 percent of the vertical frame, generous empty space at the top and bottom." Details of that safe-area discipline for still-to-video conversion are worth getting right the first time.

1:1 — profile pictures and grid uniformity. Square is the safe fallback when you do not know where an image will end up, and it is the right choice if you want a grid that reads as uniform. It is a poor choice for feed impact, since it occupies less vertical space than 4:5.

16:9 — website headers, YouTube thumbnails, X. Landscape reads as editorial and web-native. Worth noting: 16:9 is the worst possible ratio for a mobile feed and the best for a page banner, so decide which one the image is for.

3:2 and 2:3 — the photographic ratios. These are the native ratios of 35mm film and full-frame sensors, which means they are what actual photographs look like. If your goal is for an image to read as a photograph rather than as content, 3:2 landscape or 2:3 portrait carries a subtle authenticity that 4:5 does not. Use these for portfolio and about-page images.

21:9 — cinematic, and mostly a trap. Ultrawide looks striking in isolation and is unusable in almost every feed. Reserve it for site hero banners where you control the container.

The Platform Table

Format is destination, then ratio, then the tier to generate at, then how it displays:

- Instagram feed — 4:5, generate at 2K, displays 1080x1350
- Instagram Story and Reels cover — 9:16, generate at 2K, displays 1080x1920
- Instagram profile picture — 1:1, generate at 1K to 2K, displays as a 320px circle
- TikTok — 9:16, generate at 2K, displays 1080x1920
- YouTube Shorts still — 9:16, generate at 2K, displays 1080x1920
- YouTube thumbnail — 16:9, generate at 2K, displays 1280x720
- LinkedIn profile photo — 4:5 or 1:1, generate at 2K, displays as a circular crop
- LinkedIn feed image — 4:5, generate at 2K, displays 1200x1500
- X post — 16:9, generate at 2K, displays 1600x900
- Website hero — 16:9 or 21:9, generate at 4K
- Portfolio or about page — 3:2 or 2:3, generate at 2K to 4K
- Print at letter or A4 — 3:4 or 4:5, generate at 4K for 300 DPI

Resolution: Generate Above, Never Below

The rule is simple and there is no exception to it: generate larger than the display size, never smaller.

Downscaling an image throws away pixels, and the ones that remain keep their texture — the pore detail, the film grain, the fabric weave you worked to get into the prompt all survive the trip. Upscaling invents pixels, and invented pixels are smooth. Every upscale takes an image that reads as photographic and pushes it back toward the plastic look described in fixing the plastic skin problem.

Concretely: Instagram displays a feed image at 1080 wide. Generating at 1K gives you roughly that, with no margin for the platform's own recompression. Generating at 2K gives you double, which downscales cleanly and holds detail through compression. That is the entire argument for 2K as the default.

4K earns its cost in three cases: print, a website hero that will be viewed on a large display, and any image you might crop into later. Otherwise it is money spent on pixels that get thrown away on upload.

Compression, and Why Your Grain Disappears

Every platform re-encodes what you upload. That re-encoding is lossy, and it is particularly hostile to two things: fine grain and smooth gradients.

This produces a frustrating outcome. You prompt for "subtle film grain," get exactly that, upload it, and the grain either vanishes into the compression or turns into blotchy artifacts in the shadows. Three mitigations actually help:

Upload at 2K, not 1K. More source detail means the encoder has more to work with and throws away proportionally less of what you care about.

Ask for slightly more grain than you want. Compression eats some of it. Prompting for "fine, visible 35mm film grain" survives better than "very subtle grain," which often survives as nothing.

Avoid large areas of smooth gradient in dark regions. Night skies and unlit walls are where compression banding shows worst. A little texture in those areas — "faint noise in the shadow regions" — masks the banding.

Composing for the Crop You Cannot Control

Two crops happen to your images no matter what you do.

The circular profile crop. Every platform masks profile pictures to a circle. Anything in the corners of the source image is gone. Center the head horizontally, put the eyes on the upper-third line, and leave a comfortable margin above the head. If the shoulders are clipped at the frame edges in the source, the circle will make it look cramped.

The feed preview crop. Instagram's grid view shows a 1:1 crop of a 4:5 post, taken from the center. So a 4:5 image with the subject positioned near the bottom of the frame will look fine in the post and beheaded in the grid. Keep the subject within the central square of a 4:5 composition, and use the top and bottom margins for environment rather than anything essential.

You can prompt for both directly: "subject's head centered horizontally, eyes on the upper third line, generous margin above the head, all essential subject matter within the central square of the frame."

A Workflow That Avoids All of This

The way to stop fighting formats is to decide the destination before generating, then treat ratio and resolution as part of the prompt rather than as settings you adjust afterward.

For a series that needs to appear in several places — a feed post, a Story, and a website — generate each ratio separately using the same locked prompt, changing only the ratio and the composition clause. That keeps the subject identical across formats, which is the same discipline that keeps a face consistent across a batch, covered in getting consistent AI photos across generations.

And once the images exist, there is one more step worth building into the habit before anything gets posted: check the labeling requirements for the destination platform. The current state of that is in do you have to label AI photos.

Where PROMPTMVSTR Fits In

Every prompt in the archive includes composition language rather than leaving the frame to chance, which is what makes them reusable across ratios instead of only working at the size they were originally generated. The Virtual Photoshoot handles multi-format batches from one reference upload, and if you are building for a specific destination, AI photos for an Instagram personal brand covers what the feed rewards beyond the technical specs.