AI Image Size Calculator
Find the recommended resolution for every major AI model — SDXL, Flux, Qwen-Image, Seedream, GPT Image, Veo, and Kling — or check any custom size.
100% free · no signup · unlimited use
SDXL-family models are trained around ~1 megapixel. Stick to these bucket sizes (multiples of 64) to avoid artifacts and duplicated limbs.
| Ratio | Resolution | MP | Best for | Copy |
|---|---|---|---|---|
| 1:1 | 1024 × 1024 | 1.05 | native square | |
| 3:4 | 896 × 1152 | 1.03 | portrait | |
| 4:3 | 1152 × 896 | 1.03 | landscape | |
| 2:3 | 832 × 1216 | 1.01 | tall portrait | |
| 3:2 | 1216 × 832 | 1.01 | photo landscape | |
| 9:16 | 768 × 1344 | 1.03 | phone wallpaper | |
| 16:9 | 1344 × 768 | 1.03 | widescreen | |
| 21:9 | 1536 × 640 | 0.98 | ultrawide banner |
Resolution is a quality setting
Every AI image model has a native resolution it was trained at, and drifting far from it costs quality. Ask SDXL for 512×512 and you get mush; ask it for 2048×2048 in one pass and you get two heads. The "best sizes by model" tab lists the bucket resolutions each model family actually likes, per aspect ratio, with one-click copy. The custom checker tells you any size's true ratio and megapixels, snaps it to model-friendly multiples of 64, and suggests the nearest model-ready sizes with the same shape.
Skip the math entirely
If you'd rather not think about buckets and multiples at all, the Imageny playground applies the right resolution automatically for every model — you just pick an aspect ratio. Pair this calculator with our AI art prompt generator to get both the words and the pixels right.
Frequently asked questions
Why do AI image sizes need to be divisible by 64?
Diffusion models process images in latent blocks, and most architectures downsample by a factor of 8 or more. Dimensions that aren't clean multiples of 64 (or at least 32) get padded or cropped internally, which can cause soft details, edge artifacts, or unexpected framing. Snapping to multiples of 64 keeps you in the model's comfort zone.
What's the best resolution for SDXL?
SDXL was trained around 1 megapixel. The safest sizes are 1024×1024 (1:1), 1152×896 (4:3), 1216×832 (3:2), 1344×768 (16:9), and their portrait mirrors. Going far beyond ~1MP in a single pass tends to produce duplicated subjects; upscale afterwards instead.
What resolution should I use for AI video?
For Veo 3.1, use 16:9 at 1080p (1920×1080) for final output or 720p for faster drafts; vertical 9:16 renders at 720p. Kling supports 16:9, 9:16, and 1:1 at up to 1080p with 5 or 10 second durations. Source images for image-to-video should be at least 300px on each side.
What aspect ratio should I pick for social media?
9:16 for TikTok, Reels, Shorts, and phone wallpapers; 1:1 for profile pictures and album art; 16:9 for YouTube thumbnails and desktop wallpapers; 2:3 for Pinterest and posters. The calculator's ratio chips include the typical use for each shape.
Is this calculator free?
Yes — free, no signup, unlimited use. It's a free tool from Imageny, an AI image and video generation platform where these presets are built into the playground.