Online vs Offline Image Enhancement: Which Fits Your File?
· ClapClip Team · 6 min read
Pick the browser for one non-sensitive photo, and the Windows app when the file must not leave your machine or you have a folder to get through. The browser uploads the image, processes it on a GPU worker and hands back a PNG; the app runs the same kind of models on your own graphics card. Neither can guarantee that the detail it reconstructs matches the scene that was photographed.
Key takeaways
- The browser accepts one JPG, PNG or WebP at a time, up to 6 MB, 8 megapixels and 8192 px per side. It returns a PNG at 2× or 4×, with no account and no usage quota.
- 4× fits only when the source is 2 megapixels or less. The output cap is 32 megapixels, and 4× produces sixteen times the source pixels.
- The Windows app also reads TIFF, queues a folder and keeps the file on your disk. It asks for a GPU rather than an upload.
- Both paths predict detail rather than recover it. The comparison with the original is the only honest check.
Which path fits your file?
The decision is mostly about where the pixels go and how many files you have. The table below makes that trade explicit.
| Browser on enhance.place | ClapClip for Windows | |
|---|---|---|
| Where the file goes | Uploaded to a GPU worker | Stays on your disk |
| Accepts | JPG, PNG, WebP | JPG, PNG, WebP, TIFF |
| Scale | 2× or 4× | 2× or 4× |
| Output | PNG | PNG |
| Files per run | One | A queue, with one settings pass |
| Limits | 6 MB, 8 megapixels, 8192 px per side, 32 megapixels out | Your GPU and disk |
| Best for | A quick single-photo test | Private media, batches, repeated work |
The browser tool is free to use and asks for no account. The real difference is where the pixels are processed and what you can put in the queue.
Rule: if the file cannot leave your machine, stop comparing and use the app; if it is one non-sensitive photo, the browser is the simpler path.
Why does 2× fit more often than 4×?
Scaling multiplies both axes. A 1000 × 750 image is 2000 × 1500 at 2×, which is four times the pixels, and 4000 × 3000 at 4×, which is sixteen times the pixels.
The output cap turns that into a hard rule: 32,000,000 ÷ 16 = 2,000,000, so in the browser, 4× only works when the source is at most 2 megapixels. A 1920 × 1080 frame is 2,073,600 pixels and is already over the line, and most phone photos are well over it.
So the realistic browser workflow is:
- Small source, clean detail, need a large copy: 4× is available.
- Anything from a modern phone: 2× is what the caps allow, and it is usually the better-looking result anyway.
- Need more than 2× on a large original: that is a desktop job, or a two-step workflow you should not expect to be faithful.
Rule: if the source is over 2 megapixels, use 2×; if it is over 8 megapixels, the browser will not take it at all.
What should you check before keeping a result?
- Edges. Bright or dark rims around hair, buildings and text are sharpening halos.
- Faces. Compare eyes, teeth and skin at 100%. A plausible change is still a change.
- Text. Read every character that matters. Do not rely on enhancement to recover a document, a serial number or a plate.
- Texture. Fabric, foliage and gravel are where invented patterns show up first.
- Same-size comparison. Look at both versions at the same displayed size. A larger file can look better simply because it is larger.
The tool's own before-and-after comparison exists for exactly this, and it is worth using before you download rather than after. Waiting until the file is on your disk means comparing from memory instead of side by side.
Rule: if a face, a letter or a serial number changes, reject the result even when the overall image looks better.
Which error are you getting?
| Message | Cause | What to do |
|---|---|---|
| "Choose a valid JPG, PNG or WebP image." | The browser accepts three types, and a HEIC or TIFF is not one of them | Convert it, or use the Windows app, which also reads TIFF |
| "The file is too large. Choose an image under 6 MB." | Encoding size, not dimensions | Re-export at a lower quality, or crop |
| "This image exceeds the online size limit." | Over 8 megapixels, or a side over 8192 px | Shrink the image; compressing it will not change the pixel count |
| "This image is too large for the selected scale." | The output would pass 32 megapixels | Choose 2× |
| "The online queue is full. Try again shortly." | A worker is busy | Retry shortly; this is an admission limit, not a count against an account |
| "Enhancement failed." | The job used its retries without completing | Try a smaller image, or move to the desktop app |
None of these are enhancement quality problems. They are all admission checks, and each one names the limit it hit.
Rule: if the message names a size limit, resize; if it names the scale, choose 2×; if it names the queue, retry rather than editing the file.
How do you run each path step by step?
Browser:
- Drop in a JPG, PNG or WebP, one at a time.
- Choose 2× unless the source is small and the extra size matters.
- Choose Photo for camera images and Artwork for drawings and screenshots.
- Compare before and after at the same size, then at 100%.
- Download the PNG, and keep the original.
Windows app:
- Open the image, or select a folder for a batch.
- Pick General, Anime or Portrait. Portrait is for faces filling the frame.
- Set 2× first. Step up to 4× only where the comparison supports it.
- Check the preview on a representative file before queueing the rest.
- Export to a separate output folder and keep the originals.
If the source is soft rather than small, fix the softness first with the blurry photo workflow, because enlarging a soft photo mostly enlarges the softness. The offline enhancer page covers the local case in more depth, batch enhancement covers folders, and the image enhancer page is the reference for what each mode is tuned for.
Rule: if the source is soft, treat the softness before you raise the scale; resolution cannot restore detail that the capture never held.
What's the bottom line?
The browser wins on convenience and loses on privacy, file size and batch. The app wins on all three and costs you an install, a GPU and your own time on failed jobs. Neither removes the need to compare the result with the original, so keep the original and try a photo in the browser or download the Windows app to decide on your own files.