2× vs 4× Upscaling: Which Should You Choose?

Choose 2× for a moderate, natural-looking enlargement and 4× for small or heavily compressed sources that need substantially more pixels. ClearPix runs Real-ESRGAN at 4× internally; the 2× option is downsized from that result. More pixels do not guarantee better detail, so compare the image rather than choosing 4× automatically.

Start with the source, not the destination

The right setting depends on what information survives in the input:

Source	Start with	Why
Clean image around 1500 px	2×	A moderate increase usually looks more natural.
Small thumbnail or compressed social image	4×	There is more room to rebuild edges.
Old, badly degraded scan	2× first	Heavy degradation can look exaggerated at 4×.
AI art for print	4×	A 1024 px source becomes 4096 px.
Screenshot for a slide	2×	UI edges usually need less enlargement.

Super-resolution predicts plausible detail. It does not recover the exact pixels that were discarded by resizing or compression. If an original is available, use it instead of repeatedly enlarging a smaller copy.

Steps

1. Open the ClearPix image upscaler and drop in the highest-quality source you have.
2. Choose 2× for a gentle improvement or 4× for a small, heavily compressed image. Export both if the difference is not obvious.
3. Compare edges, faces, texture and text at normal viewing size. Lower sharpening if the result looks crunchy or painted.
4. Download the version that looks more faithful, not simply the one with more pixels.

What to watch for

• Faces and skin: aggressive sharpening can produce a waxy or painted appearance.
• Text and UI edges: useful when the original characters are readable, but not a substitute for OCR.
• AI artifacts: malformed hands and garbled text remain malformed and may become more obvious.
• Film grain and painterly texture: extra grain can look like noise after upscaling.

Input limits are device-dependent: up to 2048 px on the longest side with WebGPU, or 1400 px with the WebAssembly fallback. Larger images can be shrunk first. For evidence from the current pipeline, see the public benchmark and downloadable data.