SDXL native resolution picker
Stable Diffusion XL produces its cleanest results at the resolutions it was actually trained on. This picker lists every official SDXL training bucket, each landing near the model’s one-megapixel target, with exact width, height, aspect ratio and megapixel count — and one click copies the dimensions.
Why native resolutions matter
SDXL was trained with aspect-ratio bucketing: instead of one fixed square,
the data was grouped into a fixed set of width/height pairs all close to
1024×1024. Generating at one of those buckets means the model has seen that
canvas shape and frames the subject correctly.
Stray far from a trained bucket and the familiar failure modes appear:
- Too small — soft, low-detail output.
- Too large in one pass — duplicated heads, twin subjects, repeated patterns.
- Odd aspect ratio — stretched bodies and awkward cropping.
The fix is simple: pick a native resolution here, generate, then upscale afterward if you need a bigger image.
Tips for picking a resolution
- Square (1024×1024) is the safest default and the most thoroughly trained.
- Portrait (832×1216, 896×1152) suits people and posters.
- Landscape (1216×832, 1344×768) suits scenery and wide shots.
- Upscale, don’t oversize. Generate native, then run hires fix or a dedicated upscaler to reach print resolution without artifacts.