Upscale an image to 2x or 4x with the AI image upscaler, then inspect detail and artifacts at the size you will use.
A real example: a reduced copy of a watch photo beside its 4x upscale from Potto. Your own output may redraw fine detail.
The same illustration before and after a 2x upscale in the AI image upscaler. Both are shown at the same display width, then inspected at full size. Look beyond sharpness.
A crisp-looking letter can change its meaning, and a softened diagonal can become an invented outline. Compare actual characters, repeated parts and fine edges with the source. More pixels can make interpreted detail easier to see without making it correct.
A plain wall or soft background may pick up texture it never had, and dark regions can show invented structure. Decide whether those changes are harmless in the intended use. When two versions both work, keep the one with fewer unwanted changes.
Place the output at its real display width, then inspect important areas at full size. Each view finds problems the other misses. Compare source and result at matching sizes, because one zoomed in and one zoomed out gives a false impression of improvement.
The scale follows the destination, and each kind of image needs a different review. Decide what must stay dependable, then check it first.
8001,600 px
Ask the photographer or supplier for a better original first; a real capture holds more detail than an enlarged preview. If you must enlarge, check outlines, buttons, seams, fasteners and printed claims, because a sharper texture is not automatically the correct one. Keep source and result under different names so a derivative is never mistaken for the original.
6002,400 px
Look for the original scan or print before uploading. After enlarging, compare eyes, mouth, hairline and jewelry, because a plausible face is not necessarily the same face. Enlargement does not restore faded colors or remove scratches, and generated facial detail is not historical evidence. Keep the unaltered scan beside the viewing copy.
1,2002,400 px
An image upscaler fixes the size, not the anatomy, lettering or geometry of the source. Inspect fingers, railings, reflections and repeated pattern units, since enlargement makes existing problems easier to see. If the concept itself needs revision, return to the creation step and keep revision and enlargement as separate decisions.
Three steps in the panel above: upload, choose 2x or 4x, then compare with the source.
Add a JPG, PNG or WebP image up to 10 MB, using the earliest copy you have rather than a chat or screenshot version, since repeated lossy saves leave blocks that a model may read as texture. Paid plans can add up to 10 images in one batch; each becomes its own task. Keep an untouched copy.
Measure the space the image must fill first. The output long edge is capped at 2,048 pixels for 2x and 4,096 for 4x, so a 1,500 by 1,000 source cannot reach 3,000 by 2,000 at 2x and is scaled to fit. A source already at a cap cannot use that option. Read the output size, not the label.
2x costs 2 credits per image and 4x costs 6, shown on the Generate button before you confirm. Compare the source and result at matching display size, then inspect faces, lettering and edges at full size. Download the accepted image, confirm its actual dimensions and place it without stretching.
An image upscaler adds pixels to the frame you already have. These groups show where that helps and where to look twice.
01
Measure the space the image has to fill before you pick a scale. The output long edge is capped at 2,048 pixels for 2× and 4,096 pixels for 4×, so a request that would pass its cap is scaled to fit and the real factor is smaller. Read the output size, not the label.
02
The model interprets the detail it can see, so outlines, seams and surfaces look cleaner at full size. A sharper texture is not automatically the correct one, so check buttons, fasteners, dials and printed claims on a product photo against the source before it goes into a listing.
03
The aspect ratio is preserved, so upscaling never crops or stretches the source, and it does not add the scene around it. If the layout needs space outside the frame, that is a different task from enlargement. Paid plans can add up to 10 images in one batch, and each image is its own task.
04
Old prints are a common reason to upscale an image: a small scan becomes large enough to view on a screen or reprint. Scan the original at the highest resolution you can before uploading, because a better scan beats a bigger enlargement, and treat any sharper facial detail as interpreted, not recovered.
An image upscaler adds pixels to the same frame. These tools cover what else an image can need.
What an image upscaler does, how to choose 2x or 4x, and what to check when you upscale an image.
An image upscaler increases image pixel dimensions while processing the visible detail. Potto AI offers 2x and 4x options with output limits. The output remains a raster image in the source’s aspect ratio, and newly interpreted detail is not proof of the original appearance.
No. The output long edge is capped at 2,048 pixels for 2x and 4,096 for 4x. A request that would exceed its cap produces a smaller actual factor, and a source already at or beyond the option’s limit cannot use it. In a batch, one incompatible image can make the chosen scale unavailable for the group. Read the source and output dimensions before choosing.
The aspect ratio is preserved: upscaling does not crop or stretch the source. Choose the scale from the space the image must fill, not because 4x sounds stronger. An enlarged image that would exceed a limit is scaled down to fit; a tier already reached by the source is unavailable. If the layout needs extra space at the edges, that is a different task from enlargement.
No. Upscaling enlarges the existing frame; it does not add the surrounding scene, recover lettering that cannot be read or establish the exact detail of a small face. A result can look convincingly sharp while changing a number, a label or a facial feature, so compare every word and feature that matters with the source. For a screenshot or diagram, retake it at the needed size or keep the words as editable text. Flaws in generated art are enlarged along with everything else.
You can upload images and try the 2x and 4x options before signing in. Upscaling and downloading need an account; the sign-in dialog keeps your images in place.
Upload JPG, PNG, or WebP up to 10 MB per image; the output keeps the upload format. Use the earliest copy you have, because repeated saving and messaging compression add blocks that a model may read as texture. Paid plans support up to 10 images in one batch. Each image has an independent task and credit outcome, so a failed image does not block completed results. Download all packages the available results in a ZIP.
2x costs 2 credits per image and 4x costs 6, and the quote on the Generate button updates when you change the scale. In a batch, each image is charged on its own, and a failed image releases only its own reserved credits. Plans and credit packs are explained on Pricing.
Yes, on a paid plan; images made on the free plan are for personal use. Potto adds no watermark to downloads. Upscale only images you own or may use, and check enlarged labels, numbers and faces against the source before a print or a product page, because added detail is interpreted, not recovered.
Upload an image, pick 2x or 4x in the AI image upscaler and compare the result with the source before you download.
