Before you let AI clean up a photo

Upscaled a photo with AI and now it's someone else?
Try upscaling that doesn't redraw instead

Ask ChatGPT or Gemini to "make this blurry photo clearer," and it really does come out crisp. But look closely and sometimes the face is a little different, the words on a sign have changed, or something that wasn't there has appeared. That's because generative AI isn't "enlarging" your photo. It's "redrawing" it, using your photo as a reference. Here's how to choose an upscaling method when you don't want your memories or your record photos to turn into a different photo.

Doesn't redraw your photo No image upload No sign-up, no usage limits No watermark, free
THE SHORT VERSION

If you want your photo sharper "just as it is," choose upscaling that doesn't redraw. Generative AI redraws a new image using your photo as a reference, so faces, text and the number of objects can change. Shaki enlarges your photo while AI fills in the fine detail, so the composition and people's faces stay based on the original photo. "Remaking" a photo, like changing the background or removing objects, is where ChatGPT and Gemini shine.

Regular 2x enlargement A small photo simply stretched to 2x. The whole image is blurry
2x upscale with Shaki The same photo upscaled 2x with Shaki. The faces and the text on the sign are unchanged, and the outlines are sharp

For this example, the original photo was shrunk to half its size and then brought back up to 2x. In the Shaki version, the two people keep the same faces, and even the frames of the glasses and the text on the sign are sharp.

1

Generative AI doesn't "enlarge" your photo

It redraws a new image using your photo as a reference. That isn't photo upscaling.

Both get called "upscaling," but what happens inside is completely different.

Upscaling that doesn't redraw (super-resolution) = Shaki
Original photoThe small photo you have
→
Enlarge while filling in detailAI estimates and adds back blurry outlines and textures
→
The same photo, bigger and sharperComposition, faces and objects stay as they were
Generative AI (ChatGPT, Gemini, etc.)
Original photoThe small photo you have
→
Read it as a referenceWorks out what's in the photo
→
Redraw it from scratchLooks almost the same, but the details become a different photo

Super-resolution is AI that has learned, from many pairs of "blurry photos" and "sharp photos," how blur relates to the way things really look. Based on the arrangement of pixels in your photo, it fills in only the missing detail. What appears where is still decided by your original photo.

Generative AI, on the other hand, reads your photo as "reference material" and builds the image again from scratch. The result looks very much like your photo, but under the hood the whole photo is being redrawn, so even parts that didn't need redrawing can change.

Think of it this waySuper-resolution is like "carefully enlarging and restoring a photo." Generative AI is like "a skilled artist copying the photo by hand." However good the copy is, it isn't the original photo itself.
2

5 changes that often happen when you upscale with generative AI

Be especially careful with family photos, record photos and product photos
  • 1Faces start to look like someone elseThe eyes, mouth and face shape can shift little by little until you get a face that "looks similar, but isn't them." In family photos, ID photos or profile pictures, that's hard to ignore.
  • 2Text and logos changeText on signs and labels, especially non-English text and the shapes of logos, can get garbled or rewritten into different characters.
  • 3Things that weren't there appearFences, patterns or small objects that weren't in the original can get drawn in. For photos you want to keep as a record, that changes the "facts."
  • 4The number or shape of things changesSmall details, like the number of windows, doors or lights, or the shape of a vehicle, can come out different.
  • 5The composition or colors changeThe framing can shift slightly, the image can get cropped without you asking, or the colors and mood can change.

Generative AI gets better every year at "staying close to the original photo," and this doesn't happen every time. But because it works by redrawing, some variation is unavoidable.

A few more things to keep in mind beyond the photo itself

  • A mark saying it was made or edited with AI: Depending on the service, an invisible digital watermark or information indicating AI generation is embedded in the image. On free plans, a visible mark may be added as well.
  • Your photo goes to the cloud: The image is sent to the service's servers. Depending on your settings, it may be used to improve (train) the AI.
  • Usage limits, sign-up and fees: There may be a cap on how many times you can use it, or you may need to log in or pay for a plan. It also takes effort to keep rewording your request until you get what you want.
3

What Shaki can do: upscaling that doesn't redraw

Keep the photo as it is, just sharper

Shaki is a free AI image upscaler that runs right in your browser. Using super-resolution AI (Real-ESRGAN), it enlarges photos 2x or 4x while filling in fine detail.

What stays the sameComposition and framing, the people and objects in the photo, facial features, the shapes of text and logos, where the colors are
What AI fills inThe "missing detail," such as blurry outlines, the texture of hair and fabric, and fine lines
  1. Open Shaki and choose an image (PNG, JPEG or WebP)
  2. Pick a scale (2x or 4x) and upscale
  3. Compare before and after, then download
TOOL USED
Shaki
Sharpen photos 2x or 4x with AI, without redrawing them
FreeNo uploadPNG / JPEG / WebP
Open Shaki

Your photo never leaves your device

Shaki does everything, including the AI processing, inside your own device (your browser). Your images are never sent to a server. You can use it with peace of mind, even for images you'd rather not show anyone, like family photos or work documents.

No sign-up, no usage limits, no watermark

There's no account to create and no login. You can use it for free on as many images as you like, and no watermark or AI-generation mark is added to the finished image. If you hold the rights to the original image, you're free to use the result for personal or commercial purposes.

4

What Shaki isn't good at

Not redrawing comes with its own limits
Left: a sign in the original photo. Right: the photo shrunk to a quarter of its size, then enlarged 4x with Shaki. The small English letters at the bottom have lost their shape and look like different letters
Left: the original photo. Right: the photo shrunk to a quarter of its size, then enlarged 4x with Shaki. If the source is too small, the shapes of small text can break down a little.
  • Photos that are too small or badly blurred: There's a limit to how much detail can be filled in. Small text won't necessarily become readable, and its shape may break down.
  • It won't rebuild faces for you: Because it doesn't redraw a face as a different face, badly blurred faces may not come out fully clear.
  • Size and format rules: Input up to 1920px on the long side, output up to 3840px on the long side (about 4K). The scale is 2x or 4x.
  • Large images are heavy on phones: For images 1024px or larger on the long side, we recommend using a PC.
Key pointIf you want it to "look great no matter what, even if the missing parts are made up," generative AI may be the better fit. Staying true to the original photo or looking impressive: it's one or the other.

Which to use when: a quick guide

What you want to doBest pickWhy
Sharpen family photos or photos of peopleShakiFaces aren't redrawn, so no one turns into someone else
Make old photos or old phone photos easier to seeShakiDoesn't change the facts of the record, like what's in the photo or how many people
Use product photos at a larger sizeShakiEnlarges without changing shapes, logos or colors
Enlarge screenshots, documents or diagramsShakiDoesn't rewrite text (very small text has its limits)
Upscale images you'd rather not show anyoneShakiYour image is never uploaded
Enlarge illustrations or anime imagesShakiFills in detail in illustrations and anime, too
Swap the background or remove objectsGenerative AIAdding and removing things is what generative AI does best
Change clothes, poses or art styleGenerative AI"Remaking" the image is the whole point
Colorize a black-and-white photo or fill in parts that weren't capturedGenerative AI (check the result)Anything not in the original has to be "invented"
Using bothYou can also remake an image with generative AI, then enlarge it with Shaki as the final step.

FAQ

Will upscaling with Shaki give me exactly the same photo as the original?
The composition, the people and objects in it, and their facial features stay as they were in the original. However, AI fills in the fine detail that's missing when the image is enlarged, so it isn't identical down to every single pixel. Think of it this way: it doesn't redraw, so it won't become a different photo.
Should I stop upscaling with ChatGPT or Gemini?
It depends on your goal. Generative AI is great at remaking an image: swapping the background, removing objects, changing the mood. When you want to sharpen a photo just as it is, upscaling that doesn't redraw is the better fit.
Can it sharpen photos where the face is blurry?
Outlines and eyes get sharper, but it won't redraw the face as a different face. Because of that, badly blurred faces may not come out fully clear.
Can I use images upscaled with Shaki commercially?
If you hold the rights to the original image, you're free to use it for personal or commercial purposes. The copyright of the original image stays with its original owner, and Shaki holds no rights to your images.
Does it work on phones?
Yes. It runs in the browser on iPhone and Android, with no app to install. The AI model (about 64MB) loads the first time you use it, and processing uses a lot of memory, so for images 1024px or larger on the long side, we recommend a PC.
Does it work on illustrations and anime images?
Yes. Not just photos: it can also enlarge illustrations and anime images while filling in fine detail.

* The example photos are from Pexels, whose license allows editing and commercial use, and were processed with the same AI model as Shaki. Generative AI features and free tiers can change, so this article reflects things as of October 2026.

About the author

Kentaro Kawahara

Kentaro Kawahara

CEO of SHIN-YU LLC / Motion designer

Last updated:

A motion designer for 20 years, working mainly on video ads. He runs the YouTube channel "TORAERA," which offers After Effects tutorials, and has appeared on Adobe's official show "Adobe CC Dojo." He also supervises and teaches the video creator course at ZEN Study (formerly N Yobiko).

Books

Shaki, the tool featured in this article, is developed and run by SHIN-YU LLC, a company based in Japan. It processes your images right in your browser and never sends them to a server. No sign-up is needed, and it's free to use because it's supported by ads.

Sharpen your photos without redrawing them

Shaki is free, with no sign-up. Your images are never uploaded.

Open Shaki