The Quality vs Size Trade-off
When you resize an image, you are fundamentally changing its pixel content. The result depends on the resampling algorithm your tool uses – and whether you are making the image smaller (downscaling) or larger (upscaling).
Getting this wrong produces images that are blurry, pixelated, or noticeably degraded. Here is what you need to know to do it right.
Downscaling: Making Images Smaller
Downscaling is the more common case and the more forgiving one. When you reduce an image's dimensions, multiple source pixels must map to fewer output pixels – this process is called resampling or interpolation.
Resampling algorithms
Lanczos – the gold standard for downscaling photographs. It produces sharp, high-quality results by analyzing a larger area of surrounding pixels. Slower to compute but worth it for final output.
Bicubic – slightly faster than Lanczos, excellent for most photo downscaling tasks. The default in Photoshop's "Image Size" dialog.
Bilinear – faster but softer result. Acceptable for web thumbnails where speed matters more than ultimate sharpness.
Nearest-neighbor – zero interpolation; just picks the nearest pixel. Correct for pixel art (preserves hard edges) but terrible for photographs (pixelated result).
The multi-step technique
For large reductions (reducing an image to less than 50% of its original size), doing it in a single step can lose detail. Instead, reduce in multiple steps of ~50% each. The final result retains more midtone detail. Most modern tools do this automatically, but it is worth knowing if your results look unexpectedly soft.
Sharpening after downscaling
Downscaling inherently softens an image slightly. Apply a mild unsharp mask or sharpening filter after resizing to recover crispness. A radius of 0.5–1px and amount of 80–120% is a good starting point for web images.
Upscaling: Making Images Larger
Upscaling is the hard case. You cannot invent detail that was not in the original. Any upscaling algorithm must make educated guesses about what the missing pixels should look like.
Traditional upscaling (bicubic)
Simple bicubic upscaling produces a blurry result because it interpolates smoothly between existing pixel values. Acceptable for small increases (110–150%), but anything more than 2× will look visibly soft.
AI upscaling
Modern AI-based upscaling tools (Topaz Gigapixel, Adobe Super Resolution, Stable Diffusion with upscale models) use machine learning to add plausible detail during upscaling. They produce dramatically better results than traditional algorithms for 2–4× upscaling of photographs. The output can look nearly indistinguishable from a native high-resolution photo.
When you must upscale
The realistic limit for traditional upscaling without visible quality loss is about 120–150% of the original dimensions. If you need to print a small web image at large size, accept that you need either the original higher-resolution file or an AI upscaler.
Format Considerations When Resizing
If your source is a JPEG, every round of open-edit-save adds compression artifacts on top of existing ones. When resizing a JPEG:
- Resize to the target dimensions
- Save at quality 85–92% for web, or 95%+ for print
- Do not save intermediate steps as JPEG – use PNG as your working format if you need to make multiple edits
Resizing for Specific Targets
Web images: Resize to the actual display dimensions (no larger). Serving a 3000px image where it displays at 800px wastes bandwidth and slows page load. Add srcset to serve different sizes at different viewport widths.
Social media: Each platform has optimal dimensions. Twitter cards: 1200×628. Instagram square: 1080×1080. LinkedIn banner: 1584×396. Always check current specs as platforms update them.
Email: Keep width under 600px and total message image weight under 200KB. Many email clients block images by default anyway.
Resizing in Your Browser
The Image Resizer on this site resizes images directly in your browser with no server upload. You can specify exact dimensions or a percentage, maintain aspect ratio, and download the result immediately.
Key Rules
- Always start from the highest-resolution original you have
- For downscaling, use Lanczos or Bicubic – never nearest-neighbor for photos
- Apply light sharpening after downscaling
- Do not save JPEG repeatedly – use PNG as your working format
- For upscaling beyond 150%, use an AI-based tool