How to remove JPEG artifacts from an image

What "removing" JPEG artifacts actually means

You cannot recover information that JPEG threw away. Once an image is compressed at a low quality setting, the original pixel values are gone. Every tool in this guide does the same thing under the hood: it tries to hide the damage by smoothing block edges, suppressing ringing, and filling in plausible detail. The output is always a guess.

That said, some guesses are much better than others. The right tool depends on what kind of artifacts you have and how much patience you have.

If you have not read it yet, the companion explainer on what JPEG artifacts are covers the four types (blocking, ringing, color smearing, generational loss) so you can identify what you are fighting. For preventing artifacts in the first place, the quality settings guide covers which numbers actually matter when you save.

A quick decision tree

The rest of this post walks through each option in detail.

Photoshop: the built-in filter

Photoshop has a Reduce Noise filter that includes a "Remove JPEG Artifacts" checkbox. It is not the strongest option on this list, but it ships in every Creative Cloud install and works on huge files without crashing.

Steps:

1. Open the image in Photoshop.
2. Duplicate the layer (Cmd+J / Ctrl+J). Always edit on a copy.
3. Filter > Noise > Reduce Noise.
4. In the dialog, check Remove JPEG Artifact. Set Strength around 6, Preserve Details around 50%, Reduce Color Noise around 30%, Sharpen Details around 25%.
5. Toggle the Advanced view if you want to denoise specific channels (the blue channel usually has the worst chroma noise).
6. Click OK, then add a Smart Sharpen pass at low radius if the result is too soft.

A second technique uses Camera Raw: Filter > Camera Raw Filter, then push the Noise Reduction sliders. Camera Raw's denoise model is newer and generally handles ringing better than the legacy Reduce Noise filter.

Adobe Photoshop Elements has a more aggressive JPEG Artifacts Removal preset under Enhance > Adjust Sharpness workflows; if you have Elements rather than full Photoshop, that is where to look.

GIMP and free desktop options

GIMP does not have a JPEG-specific filter, but the G'MIC plugin covers everything Photoshop does and more. Install G'MIC, then look under Filters > G'MIC-Qt > Repair > Smooth [anisotropic] or Iain's JPEG Artifact Removal. Iain's filter was written specifically for this case and is the best free option for blocking and ringing.

For Linux users who prefer command-line tools, ImageMagick has a -enhance operator and various blur/sharpen combinations, but it is generally weaker than G'MIC.

Online tools

Online tools are convenient but inconsistent. The free tiers often cap resolution at 1024px or 1500px, which makes them useless for serious photo work.

Quick comparison of the ones worth trying:

Treat the free tiers as previews. If a paid plan is worth it, run a one-month subscription and batch your archive through it.

AI models for serious cases

The state of the art for JPEG artifact removal is AI super-resolution / restoration models. They were trained on pairs of (clean image, JPEG-damaged version), so they have learned what specific artifact patterns "should" look like clean.

FBCNN (Flexible Blind Convolutional Neural Network) is the open-source standard and runs free on Hugging Face. Upload a single image, get a cleaned version. The free Space has a queue and a resolution cap, but the model is also available to run locally if you can install PyTorch.

Topaz Photo AI is the commercial benchmark. It bundles JPEG artifact removal with denoise, sharpening, and upscaling in one workflow. A one-time license is around $200; there is a free trial that watermarks the output. It is the best single tool for heavily-damaged photos, particularly portraits.

waifu2x was originally designed for anime-style art, but the JPEG-noise-reduction mode handles real photos well too if the artifacts are severe enough that detail is mostly gone anyway. Free at waifu2x.udp.jp or self-hosted.

Across all of these, expect 30 to 90 seconds per image. Quality scales with how much patience you have and how big your input is.

When to stop trying and re-export

The fastest fix is the one most people skip: find the original. If the image lives in your camera roll, a photo library, a RAW backup, or a cloud sync, re-export it from there at a high JPEG quality (or PNG, or WebP) and replace the damaged file. Five minutes of digging beats an hour of AI denoising.

If the image came from someone else, ask. People often have a higher-quality version they sent at a low quality by accident.

The only case where AI restoration is genuinely the right answer is when the high-quality original truly does not exist: a scanned photograph, a screenshot of a screenshot, or an image pulled from a third-party feed. Even then, the goal is plausibility, not accuracy.

A note on the opposite direction

If you came here looking to add artifacts to an image on purpose (memes, vaporwave-adjacent aesthetics, intentional degradation for art or privacy), that is what the JPEG Artifact Generator is for. Drop in an image, push the quality slider down, optionally add chromatic aberration and noise, download.

The companion post on deep-fried memes covers the cultural side of why anyone would want to add artifacts.

FAQ

Why does cranking the quality slider up not undo the damage?

Saving an image at higher quality only changes future generations. The compression that already happened is permanent. Saving a damaged JPEG at quality 100 just preserves the existing artifacts at higher fidelity.

Will sharpening help after I denoise?

Often, yes. JPEG artifact removal is essentially aggressive denoise, and aggressive denoise blurs detail. A small Unsharp Mask or Smart Sharpen pass after restoration recovers some of the lost crispness without bringing the artifacts back, as long as the radius is small (0.3 to 0.8 pixels).

Can AI tools recover detail that was never there?

They hallucinate detail. The output is plausible, not faithful. For portraits, this is often fine because the model has seen millions of faces and knows what a face "should" look like. For unique subject matter (scientific images, archival photographs, anything where the actual pixel content matters), be skeptical.

Does AI restoration work on screenshots and text?

Less well. Most models were trained on photographs, so they smooth screenshot artifacts away but also blur the text and UI elements you care about. waifu2x's "art" mode does better on text-heavy content than general photo models.

Is there a way to remove JPEG artifacts in the browser?

Yes, but the quality is mediocre. Browser-side ML models have to be small enough to download quickly, which caps capability. For one-off cleanups it can be enough. For serious work, use a desktop or cloud tool.

Does saving as a different format help?

It freezes the current state. Re-saving a damaged JPEG as a PNG preserves the artifacts losslessly going forward, which is useful if you plan to do further edits and do not want each save to introduce new damage. It does not fix anything that is already there.

Sources

• Jiang, J. et al. (2021). Towards Flexible Blind JPEG Artifacts Removal (FBCNN). ICCV 2021. The reference paper for the open-source removal model.
• Adobe (2024). JPEG artifacts removal filter. Photoshop documentation.
• G'MIC, the GIMP plugin used for free desktop restoration.
• Topaz Photo AI, commercial benchmark for AI photo restoration.