Free Image Upscaler
Enlarge images up to 7680x4320 for free, no credits needed. When a photo needs real detail rebuilt rather than stretched, AI Upscale 2x and 4x reconstruct it.
Two ways to upscale
- High-quality resize - free. Enlarges to 2K-8K with Lanczos resampling: clean and sharp, with no blocky artefacts. It works from the pixels you have, so it cannot invent detail - very large enlargements of a small image will look soft.
- AI Upscale 4x - 3 credits. AI super-resolution reconstructs plausible detail, so edges stay defined and texture survives. The difference shows most on faces, text, fabric and foliage.
Start with the free resize. If the result looks soft at the size you need, the AI upscale is the step up.
What it costs
Credits are a one-time purchase and do not expire while the Service is available. These are the costs the server deducts:
- 2K to 8K (up to 7680x4320) high-quality resize - free
- AI Upscale 4x - 3 credits
Your 20 free signup credits cover 6 AI 4x upscales.
Choosing a target resolution
Bigger is not automatically better. Upscaling far past what the source supports produces a large file without meaningfully more real detail.
- Print - work back from physical size at 300 DPI. A 10-inch wide print wants roughly 3000px, so 4K is usually plenty.
- Web and social - 2K is almost always enough. Larger files simply get recompressed by the platform.
- Large-format display - 6K to 8K is worth it when the image will be viewed close up at scale.
A good rule: a 4x enlargement from a sharp source beats an 8x enlargement from a soft one, every time.
Source quality sets the ceiling
Upscaling amplifies whatever is already in the file, including its flaws. Heavy JPEG artefacts, motion blur and sensor noise all get enlarged too.
If you have a choice of source, start from the largest, least-compressed version you can find. Re-saving a small JPEG several times before upscaling gives the model bad information to work from.
What upscaling cannot fix
It is worth being clear about the limits, because upscaling is often reached for when a different tool is the right answer.
- Motion blur - the detail was never recorded. Enlarging produces a bigger blur, not a sharper subject.
- Severe JPEG artefacts - blocking and ringing get reconstructed along with everything else, sometimes more visibly than before.
- Out-of-focus subjects - same problem as motion blur. Focus cannot be recovered after the fact.
- Heavy noise - grain is texture as far as the model is concerned, so it tends to be preserved and enlarged.
What it does do well is take a genuinely sharp but small image and make it usable at a larger size. That is the case it is built for.
Where upscaling belongs in your workflow
Order matters more than people expect. Upscaling is best done last, after any edits that change the content of the image.
If you remove a background, retouch an object or crop, do that at the original resolution first. Each of those operations is more accurate on unmodified pixels, and upscaling afterwards enlarges a clean result rather than baking in and then magnifying whatever the earlier step got slightly wrong.
The one exception is cropping heavily into a small image. There, upscaling first can give the later steps more pixels to work with - though a crop that severe is usually a sign the source is too small for the job.
Formats
Upload JPEG, PNG, WebP or HEIC/HEIF. Photographic results are best exported as JPEG; anything with transparency will export as PNG automatically. At 8K the difference in file size between the two is substantial.