Start with the output requirement

Write down the final width and height before choosing a mode. If a 900-pixel image already fits the destination, making it larger may add file weight without improving the use. If the destination needs a larger working image, AI 2x is the relevant test because it creates a larger output.

Choose Local Sharpen for a mild problem

Local Sharpen adjusts pixels in the browser and does not require sending the selected image to the cloud. It is a sensible first test for a slightly soft scan, a small graphic, or an image whose size is already correct. Use a restrained strength and inspect edges at 100%.

Signs it is enough

  • Letters and edges are recognizable but slightly soft.
  • The current dimensions meet the publishing requirement.
  • Privacy or offline-like handling matters more than enlargement.
  • The result looks clearer without halos or amplified noise.

Choose AI 2x for a real size need

AI 2x creates a larger WebP derivative through the cloud workflow. The browser prepares a smaller processing copy and uploads it only after you start the AI operation. High Quality and Fast are fixed modes; they are not face restoration, forensic recovery, or a guarantee of true detail.

Check these trade-offs

  • Cloud processing adds queue and execution time, especially when the worker is cold.
  • The source is resized to the documented AI limits before processing.
  • Repeated patterns, small text, hair, and faces need careful review.
  • A failed or canceled job is not a reason to submit the same image repeatedly.

A practical comparison

Suppose a product photo is 800 pixels wide but a catalogue draft needs 1600 pixels. Run Local Sharpen only if the dimensions are acceptable for the draft. Otherwise, create the AI 2x result, compare the label and straight edges against the original, and use the original or a new photograph if the model changes a fact-bearing detail.

Do not use either mode as deblurring evidence

Sharpening can increase local edge contrast. Super-resolution can estimate plausible high-frequency detail. Neither result establishes what was actually present in a severely motion-blurred or out-of-focus image. Keep the original and label an AI output as an enhancement when that distinction matters.

Decision checklist

  1. Do I need larger pixel dimensions, or only clearer edges?
  2. Can the source remain in the browser, or is cloud processing acceptable?
  3. Have I kept the original file?
  4. Have I inspected text, faces, repeated patterns, and straight edges?
  5. Does the output remain suitable at its actual display size?

For the full distinction between methods, read sharpening vs deblurring vs upscaling. For noise and softness, see which process to try first and the tool guide.