AI Image Resolution Calculator: SDXL & Flux
Turn an aspect ratio and pixel budget into exact dimensions. Pick a model to load its defaults, adjust them if needed, then compare the rounded size with an SDXL training resolution.
FreeNo signupRuns in your browser
Target area before rounding. Switching model family reloads its defaults.
Both settings stay editable. Custom keeps your current values.
Plan the text that goes with your images too. Browse free, copy-ready prompts for marketing and SEO, sales, careers and coding.
Browse free copy-ready prompts →SDXL resolutions and training buckets
SDXL was trained around a 1024 x 1024 pixel area using several aspect ratio buckets. Choose the shape that leaves room for your subject, then use a matching training size as a starting point. The uses below are composition suggestions, not limits on what each bucket can depict.
| Aspect ratio, rounded | Size in pixels | Suggested use |
|---|---|---|
| 1.00:1 | 1024 x 1024 | Square cover or centered object |
| 1.29:1 | 1152 x 896 | Gently horizontal scene |
| 0.78:1 | 896 x 1152 | Portrait with some surrounding space |
| 1.46:1 | 1216 x 832 | Landscape or tabletop arrangement |
| 0.68:1 | 832 x 1216 | Standing portrait or poster |
| 1.75:1 | 1344 x 768 | Wide cinematic framing |
| 0.57:1 | 768 x 1344 | Tall composition |
| 2.40:1 | 1536 x 640 | Panoramic landscape |
| 0.42:1 | 640 x 1536 | Narrow vertical scene |
The nearest training size is selected by absolute difference between width-to-height ratios. It is independent of your edited megapixel target. For example, a 16:9 request is about 1.78:1, while the nearby 1344 x 768 bucket is 1.75:1. The SDXL aspect ratio guide discusses choosing a shape for the intended image.
How each model family handles resolution
| Family | Calculator preset | Practical guidance |
|---|---|---|
| SD 1.5 | 0.26 MP, multiples of 64 | Trained at 512 x 512; use a hires or upscale pass for larger results |
| SDXL / Pony / Illustrious | 1.05 MP, multiples of 64 | Start with a nearby SDXL training bucket |
| Flux | 1.0 MP, multiples of 16 | Check the limits of your specific model and interface |
| Custom | Your current megapixels and multiple | Set the requirements of your own workflow |
Latent image models need dimensions divisible by their downsampling factor. Arbitrary dimensions can fail or be silently cropped. Rounding to a selected multiple addresses divisibility; it does not certify that every possible size is supported by a particular app.
Black Forest Labs documents FLUX.2 dimensions in multiples of 16, from 64 pixels up to 4 megapixels, and recommends staying at or below 2 megapixels. An extreme aspect ratio can still produce a side below that minimum. Review both dimensions, not just the area, and consult the FLUX resolution guide for that workflow.
How to use the resolution calculator
- Select the model family to load its target area and rounding multiple. Choose Custom to retain your values while removing family-specific advice.
- Pick an aspect ratio, or enter positive width and height values under Custom. A ratio describes shape, not a pixel size.
- Adjust megapixels if needed. The tool calculates width from the square root of pixel area times ratio, then divides width by ratio for height.
- Review the rounded dimensions, real ratio and real megapixels. For SDXL, prefer the recommended training size for generation.
Each dimension rounds independently to the nearest multiple, with a minimum of one multiple. This can noticeably change small or extreme shapes. Use the image prompt generator to describe the composition that fits the resulting frame. If you need an aspect ratio parameter for Midjourney, use the Midjourney prompt generator.
Generate a base image, then upscale
These are rules of thumb, not hard limits. Well above its 512 x 512 training size, SD 1.5 often duplicates subjects unless you add a hires or upscale pass. Well below its 1024 x 1024 training area, SDXL usually loses quality. For FLUX.2, Black Forest Labs recommends staying at or below 2 megapixels. The status line flags a size past roughly 0.6 megapixels for SD 1.5, under roughly 0.6 for SDXL and over 2 for Flux, so you can reconsider the base size before generating.
Decide whether you like the subject, pose and composition at the base resolution before spending time on a larger output. The 2x line doubles both dimensions, which quadruples the pixel area. It is a size reference for an upscale step, not a recommendation to generate the initial image at that larger size.
Frequently asked questions
What resolution should I use for SDXL?
Start around the area of 1024 x 1024 pixels and use an SDXL training bucket that suits the composition. This calculator recommends the listed training size with the closest aspect ratio. Its rounded calculation can differ from that recommended bucket.
What resolution was Stable Diffusion 1.5 trained at?
SD 1.5 was trained at 512 x 512 pixels. The SD 1.5 preset uses a target of 0.26 megapixels and multiples of 64. Well above that size it often duplicates subjects unless you add a hires or upscale pass.
Which image dimensions work with Flux?
The Flux preset targets 1 megapixel and rounds dimensions to multiples of 16. Black Forest Labs documents FLUX.2 dimensions in multiples of 16, starting at 64 pixels and up to 4 megapixels, with a recommendation to stay at or below 2 megapixels. Check the actual dimensions against the requirements of the Flux model and interface you use.
Why does the calculated aspect ratio change?
Width and height are rounded independently to the nearest selected multiple. That changes both the aspect ratio and the actual pixel area. The result reports the rounded ratio and megapixels so you can see the difference.
How do I calculate image dimensions from megapixels?
Multiply megapixels by one million to get pixel area. Multiply that area by the width-to-height ratio and take the square root for width, then divide width by the ratio for height. This tool rounds both dimensions and also shows a 2x upscale size.
Does a 2x upscale mean twice as many pixels?
No, doubling both width and height gives four times the pixel area. The upscale line shows the dimensions only; this calculator does not resize an image. Generate the base image in your image app before choosing an upscale workflow.
