"Improving image quality" can mean different things: increasing resolution, reducing grain, boosting perceived sharpness, or simply making sure the image doesn't lose quality when compressed. Each problem has its own solution — and mixing up the wrong approaches is the fastest way to make things worse.

In this guide you'll learn how to diagnose your image's actual problem and apply the right technique for each situation.

Why do images lose sharpness?

Before fixing it, it's important to understand the cause. The most common sources of quality loss are:

💡 Tip: Images blurred by incorrect focus during capture can't be fully recovered — not even with AI. The techniques here work for compression, resolution, and noise problems.

AI upscaling: the resolution revolution

The most effective way to enlarge images without losing sharpness is upscaling via neural networks. Tools like Topaz Gigapixel AI, RealESRGAN, and Adobe Super Resolution use models trained on millions of images to infer details that simply didn't exist in the original — instead of blurring pixels, they generate plausible textures.

The result is especially impressive on faces, fabrics, and natural landscapes. A 500×500px photo can be enlarged to 2000×2000px with convincing detail, making it suitable for printing or use in banners.

Sharpening: crisp without overdoing it

The sharpening filter works by increasing contrast at object edges — which creates the visual perception of a crisper image. The problem is that overdoing it creates white halos around elements and amplifies image noise.

The most controlled technique is Unsharp Mask, available in Photoshop, GIMP, and even online tools. The recommended settings for most images are: amount between 80–150%, radius of 1–2px, and threshold of 0–5.

Smart compression: keep quality with a smaller file

Many people confuse "quality image" with "large file." With modern formats — especially WebP and AVIF — you can have images visually identical to the original with 30–50% less file size. That's pure gain: same sharpness, faster loading.

FormatCompressionSharpness keptBrowser support
JPG (q90)GoodHighUniversal
PNGLosslessPerfectUniversal
WebPExcellentHighVery good
AVIFSuperiorHighGood (modern)

Reducing noise without blurring

Photos taken in low light show grain (noise) that reduces perceived sharpness. The classic noise reduction approach always sacrificed some sharpness, since it blurred noise along with everything else. Modern AI filters — like those in Lightroom Denoise or DxO PureRAW — can precisely separate real noise from real image information, removing one without affecting the other.

💡 Tip: After increasing resolution or applying sharpening, use the ImageTools Image Compressor to reduce file size without compromising the sharpness you worked so hard to get.

Recommended workflow

  1. Always start with the best available version of the original file
  2. Apply noise reduction before upscaling
  3. Do AI upscaling if you need to increase size
  4. Apply sharpening in moderation as a final touch
  5. Save as PNG for future edits; convert to WebP/JPG only for final delivery

Compress without losing what you improved

Use the ImageTools Image Compressor to deliver high-quality images at the smallest possible size.

Compress image now

Frequently asked questions

Which free upscaling tool gives the best quality?
RealESRGAN is the best free general-purpose model. It's available online via Upscayl (a free desktop app) and several web interfaces. For face photos, GFPGAN gives superior results.
Is increasing resolution in Photoshop with "Preserve Details 2.0" as good as AI?
No. Preserve Details 2.0 is much better than classic bicubic interpolation, but it still falls short of modern AI models for most cases. For maximum results, use specialized tools like Topaz Gigapixel or RealESRGAN.
Can video sharpness be improved with the same techniques?
Yes, but the process is more complex and time-consuming — each frame needs to be processed individually or with tools specialized in video like Topaz Video AI. For still photos, results are much faster and more practical.