How to Split a Midjourney Grid into 4 Separate Images
Split a 2×2 Midjourney or AI image grid into four separate files, preserve every pixel, handle borders, and prepare the results for publishing.
A 2×2 AI image grid is one file that contains four visual candidates. Splitting converts that composite into four independent files; it does not invent detail, upscale a chosen image, or change the composition. When the panels are equal, the task is straightforward geometry and can run entirely in the browser without uploading the source to an image-processing service.
Good results still depend on inspecting the source. Thick gutters, labels, outer frames, unequal cells, or a screenshot containing interface controls need a manual crop before equal division. This guide explains how to recognize a compatible grid, preserve odd pixels, choose an output format, and avoid publishing a quarter that still contains unwanted border fragments.
Check the estimate or cleanup workflow locally in your browser without uploading your content.
Splitting, cropping, and upscaling solve different problems
Splitting takes one 2×2 composite and extracts its top-left, top-right, bottom-left, and bottom-right quadrants. Cropping takes one image and removes outer areas to create a target aspect ratio or tighter composition. Upscaling creates a larger pixel output and may use a model or interpolation to add or estimate detail. A split result is naturally smaller than the full grid because it contains only one quarter of the source area.
Choose the operation from the file you actually have. If the four candidates already exist as separate downloads in the generation service, use those originals instead of splitting a screenshot. If you only have a clean equal grid, split it. If the chosen panel needs a square, portrait, or widescreen frame, crop the separated image afterward. If its resolution is insufficient, use an appropriate upscaling workflow after composition is final.
| Operation | Input | Result |
|---|---|---|
| Split | One equal 2×2 composite | Four independent quadrants |
| Crop | One image with extra edges | One image at a new frame or aspect ratio |
| Upscale | One image needing more pixels | A larger output, sometimes with generated detail |
Check the source before you split it
Open the image at full size and confirm there are exactly two rows and two columns. The vertical division should sit near the center, and the horizontal division should do the same. Look for white or dark gutters, rounded frames, numbers, watermarks, browser chrome, or chat-interface padding. Equal geometry cannot understand which pixels are decoration; any decoration crossing a quadrant becomes part of the exported file.
Use the original downloaded image whenever possible. Messaging previews and screenshots can be compressed, rescaled, or padded. A screenshot may also include only a visually scaled preview rather than the source pixels. Keep the original file as an archive, work on a copy, and avoid repeatedly saving JPEG intermediates because each lossy re-encode can introduce more artifacts.
- Prefer the original JPG, PNG, or WebP download over a screenshot.
- Confirm the grid is a true equal 2×2 layout.
- Inspect the center lines for gutters or labels.
- Crop browser or chat interface elements before splitting.
- Keep an untouched source file for recovery and comparison.
How equal quadrant splitting preserves odd pixel dimensions
For an even source such as 2048×2048, each quadrant is 1024×1024. Real files are not always even. If a source is 2049 pixels wide, dividing and rounding every panel to 1024 would silently discard one column. A reliable algorithm gives the left column the floor of half the width and assigns the remaining width to the right column. The same rule applies vertically.
RunAIToolkit follows this approach: top and left rectangles use the floored half, while right and bottom rectangles receive any extra pixel. Every source pixel belongs to exactly one result. The output pieces can differ by one pixel, which is correct and preferable to duplication or loss. This rule does not remove a gutter; it only divides the complete source without gaps.
| Source size | Left / right widths | Top / bottom heights |
|---|---|---|
| 2048×2048 | 1024 / 1024 | 1024 / 1024 |
| 2049×2048 | 1024 / 1025 | 1024 / 1024 |
| 2048×201 | 1024 / 1024 | 1100 / 1101 |
| 2049×201 | 1024 / 1025 | 1100 / 1101 |
Step-by-step browser workflow
Open the AI Image Grid Splitter and keep the 2×2 mode selected. Drop a supported JPG, PNG, or WebP source into the page and inspect the preview. Process the image, then review all four results in reading order: top-left, top-right, bottom-left, bottom-right. Download a single candidate or save the four outputs as one ZIP archive. The file reading, Canvas drawing, preview, and ZIP creation remain on the current device.
Do not rush from a successful download to publication. Open at least one exported file at 100 percent zoom. Check that no row or column from the neighboring panel is visible, that any grid line is acceptable, and that small text is still legible. Confirm pixel dimensions in the file inspector. If every output has half a gutter along its inner edge, return to the source and manually remove the center spacing before using equal splitting.
- Select the 2×2 split mode rather than single-image crop mode.
- Load one supported image no larger than the tool's stated limit.
- Inspect the source preview and generate four outputs.
- Check all four edges and pixel dimensions before downloading the ZIP.
- Rename final files by subject or use instead of relying only on grid order.
Choose an aspect ratio after selecting the best panel
Aspect ratio describes the relationship between width and height, not the number of pixels. Midjourney's official documentation uses the --ar or --aspect parameter and notes that the default is 1:1. It also warns that aspect ratio is not the same as image dimensions. A square can be 512×512 or 2048×2048; both are 1:1 but have very different detail and file size.
If the split panel already has the needed composition, keep it intact. Otherwise use crop mode on one separated image. RunAIToolkit offers centered 3:4, 1:1, and 16:9 crops. Center cropping never stretches the image, but it can remove important content near the long edges. If the subject is off-center or text sits close to an edge, use a manual editor that lets you reposition the crop.
| Ratio | Shape | Composition check |
|---|---|---|
| 1:1 | Square | Keep the focal point near the center |
| 3:4 | Portrait | Check the top and bottom breathing room |
| 16:9 | Widescreen | Expect substantial top and bottom removal from square art |
Protect quality, metadata, and publishing intent
PNG preserves pixel values without lossy compression but can be large for photographic art. JPEG and WebP outputs may be re-encoded by the browser, so a slight size or quality change is possible. Splitting cannot restore detail that is absent from the source. If the full grid is 2048×2048, each clean quadrant is around 1024×1024; describing it as 4K because the composite looked large would be misleading.
Canvas exports commonly omit original EXIF, location data, and some color-profile metadata. That may reduce accidental metadata sharing, but it also means the output is not a complete archival replacement. Keep the source. Before publishing, check the destination's current crop and compression preview, add descriptive alt text where supported, and verify that you have the right to use the generated image and any reference material.
- Keep the original composite and the final chosen panel.
- Use PNG when lossless pixels matter; use JPEG or WebP consciously for size.
- Do not claim that splitting increases resolution or adds detail.
- Check platform previews for another round of cropping or compression.
- Use descriptive filenames and alt text, and confirm usage rights.
Frequently asked questions
Does splitting a grid upscale the four images?
No. It extracts existing pixels. Each quadrant is smaller than the complete composite and no new detail is created.
Can the tool remove white lines between panels?
It divides the complete image into equal quadrants. A gutter is included in the results unless you crop it from the source or outputs.
What happens when the width or height is an odd number?
The extra pixel goes to the right column or bottom row, so every source pixel is preserved exactly once.
Is the image uploaded to a server?
No. File decoding, Canvas processing, previews, and ZIP creation happen in the browser on the current device.
Official sources
Product documentation, specifications, and prices can change. Recheck these pages before relying on them.