The best aspect ratios for Stable Diffusion on social media are 1:1 or 3:4 for Instagram posts, 9:16 for Stories, Reels and Shorts, 2:3 for Pinterest and 16:9 for YouTube thumbnails. Generate at a size close to what the model was trained on, such as 896×1152 or 768×1344 for SDXL and about 512 pixels for SD 1.5, then crop and resize to the platform's exact pixels.
Best aspect ratios for social media: the short answer
Pick the ratio from where the image will be posted, then generate the closest size your model handles well. Instagram keeps feed photos at full resolution between 1.91:1 (wide) and 3:4 (tall), up to 1080 × 1440 pixels. Stories use 9:16 at 1080 × 1920, Pinterest recommends 2:3 at 1000 × 1500, and YouTube asks for 16:9 video thumbnails and 9:16 Shorts thumbnails.
Stable Diffusion does not need to hit those pixel sizes directly. Generate near the model's training size in the same shape, then crop the last few pixels and resize. That keeps the composition intact instead of letting the platform crop it for you.
Aspect ratio cheat sheet by platform
Instagram feed posts: 1:1 (1080 × 1080) or portrait up to 3:4 (1080 × 1440). Instagram's help page says photos between 1.91:1 and 3:4 keep their resolution at a width of 1080 pixels, and that other ratios are cropped to fit.
Instagram Stories and other fullscreen vertical formats: 9:16. Meta's Stories specification recommends 1080 × 1920. YouTube Shorts thumbnails also use 9:16.
Pinterest: 2:3, or 1000 × 1500 pixels. Pinterest notes that Pins taller than 2:3 might get cut off in people's feeds.
YouTube video thumbnails: 16:9, with 3840 × 2160 recommended and a minimum width of 640 pixels. Podcast playlists use 1:1.
SDXL: generate in its training buckets
SDXL was trained on many aspect ratios, not only squares. The SDXL paper describes splitting the training images into buckets of different aspect ratios, each with a pixel count close to 1024 × 1024 and sides in multiples of 64. Generating at one of those bucket sizes gives the model a shape it has seen.
The buckets closest to the social ratios are 1024 × 1024 for 1:1, 896 × 1152 for 3:4 and 4:5 posts, 832 × 1216 for 2:3 Pins, 768 × 1344 for 9:16 Stories and Shorts, and 1344 × 768 for 16:9 thumbnails. Each one is a little off the exact ratio, so crop a few pixels after generating and then resize to the platform's size.
PhoneDiffusion generates on-device on iPhone or iPad, with optional cloud models. Download on the App Store →
SD 1.5: stay close to 512 pixels
Stable Diffusion 1.5 was trained at 512 × 512, and the AUTOMATIC1111 wiki warns that SD 1 and 2 models create bad images at very high resolutions. For portrait or landscape images, keep one side at 512 and stretch the other only moderately, for example 512 × 768 for 2:3. Width and height must be divisible by 8 in the Diffusers pipeline.
Do not ask an SD 1.5 model for a full 1080 × 1920 Stories canvas in one pass; that is far above its training size. Generate smaller first, then upscale. In AUTOMATIC1111, Hires. fix does that in one go: it renders at a lower resolution, upscales and adds detail in a second pass.
Crop, extend or upscale for the final size
Leave some room around the subject when you generate, because the last step is usually a crop. On iPhone, the Photos app has crop presets such as Square, Wallpaper, 9:16 and 5:7; tap the lock at the top right to unlock the ratio and drag the corners to any shape, such as 4:5 or 2:3.
To make a square image taller or wider without cutting it, extend it instead. In iOS 27 with Apple Intelligence on an iPhone 15 Pro or later, the Photos Extend tool expands the edges of a photo beyond its original frame. Then resize to the platform's pixels. Instagram keeps photos between 320 and 1080 pixels wide at their original resolution and scales wider ones down to 1080, so upscaling to 1080 wide gives the sharpest post.
If a crop changes the image more than you want, do not just raise the size and generate again. A new size usually changes the whole composition even with the same seed; see Stable Diffusion seeds explained.
PhoneDiffusion availability and article scope
PhoneDiffusion runs curated Core ML model packs (SD 1.5, SD 2.1, and SDXL) on supported iPhone and iPad devices with iOS 17 or later; local generation is private and works offline once a pack is downloaded. In the current release every local pack generates square images: 512 × 512 for the SD 1.5 and SD 2.1 packs and 768 × 768 for the SDXL packs, which are built for that one size. The SDXL buckets and custom sizes above apply to desktop tools.
The optional cloud models on paid monthly or yearly plans, such as FLUX and Seedream, also return square images, mostly 1024 × 1024. They are text-to-image only, and the text prompt is sent online to generate the image. For a 9:16 Story or a 2:3 Pin, compose with space around the subject, save the image, then crop or extend it in Photos.
Field notes
- Instagram posts: 1:1 or up to 3:4 (1080 × 1440); Stories: 9:16 (1080 × 1920).
- Pinterest: 2:3 (1000 × 1500); YouTube thumbnails: 16:9, Shorts 9:16.
- SDXL: generate in a training bucket such as 896 × 1152, 832 × 1216, 768 × 1344 or 1344 × 768.
- SD 1.5: keep one side near 512 and upscale afterwards.
- Crop, extend or resize at the end instead of generating at the exact platform size.
Try it on your phone
- Choose the platform and its ratio from the cheat sheet.
- Pick the closest size your model was trained on (SDXL bucket or about 512 for SD 1.5).
- Generate with room around the subject.
- Crop to the exact ratio in Photos, or extend a square image.
- Resize or upscale to the platform's pixel size and post.
Sources and verification
Primary and maintainer documentation checked for this guide. Product-specific details were reviewed against the current PhoneDiffusion scope.
- Instagram Help Center: Image resolution of photos you share on InstagramPhotos up to 1080 pixels wide keep their resolution when the aspect ratio is between 1.91:1 and 3:4 (height 566 to 1440 pixels at 1080 wide); unsupported ratios are cropped.
- Meta Business Help Center: Design requirements for Instagram Stories adsStories use the fullscreen vertical format: 9:16 aspect ratio, recommended resolution 1080 × 1920.
- Pinterest Business Help: Review ad specsStandard image Pins: 2:3 aspect ratio recommended, or 1000 × 1500 pixels; taller than 2:3 might get cut off in feeds.
- YouTube Help: Add custom thumbnailsVideo thumbnails 16:9 (3840 × 2160 recommended, at least 640 wide); Shorts thumbnails 9:16; podcast playlists 1:1.
- Podell et al.: SDXL paper (multi-aspect training)Section 2.3 and Appendix I: training data split into aspect-ratio buckets with a pixel count close to 1024², height and width in multiples of 64, applied as the final training stage; buckets include 1024×1024, 896×1152, 832×1216, 768×1344 and 1344×768.
- Stable Diffusion v1.5 model card: limitationsModel card listing ‘The model cannot render legible text’, a lossy autoencoder, and mainly English training captions; uses a fixed CLIP ViT-L/14 text encoder and was trained at 512×512.
- AUTOMATIC1111 wiki: Features (Hires. fix)Hires. fix renders at a lower resolution, upscales and adds detail; the wiki says SD1/2 models create bad images at very high resolutions because they were trained at 512px or 768px.
- Hugging Face Diffusers: StableDiffusionPipeline source (size check)check_inputs raises ‘`height` and `width` have to be divisible by 8’; width and height default to the UNet sample size times the VAE scale factor.
- Apple Support: Crop, rotate, flip, or straighten photos and videos on iPhonePhotos crop presets such as Square, Wallpaper, 9:16 and 5:7; unlock the aspect ratio with the lock at the top right to crop freely.
- Apple Support: Clean up, extend, and reframe photos on iPhoneWith Apple Intelligence (iOS 27 or later, iPhone 15 Pro or later), the Extend tool expands the edges of a photo beyond its original frame.