Make every pixel count
Add an image to upscale it 2×, 4× or 8×.
Drop, paste or browse. JPEG, PNG, WebP and more.
ashUp enlarges images on your device, without uploading them. There is nothing to install and no account to sign into.
Choose Lanczos for a fast enlargement, or try an AI model to create a sharper-looking result. AI predicts detail, so it can also change faces, lettering and textures. Compare the result with your original before downloading it.
You can enlarge an image to 2×, 4× or 8× its width and height, depending on the model and what your browser can handle. Your original file is not changed.
Download any result you want to keep before replacing the image, closing the tab or refreshing the page. Images and finished results are not saved as a project you can reopen later.
The scale multiplies both the width and height. At 2×, a 1,000 × 750 image becomes 2,000 × 1,500, with four times as many pixels. At 4× it has 16 times as many pixels, and at 8× it has 64 times as many.
Choose the smallest scale that gives you the dimensions you need. More pixels require more memory and usually take longer to process and save. A larger image does not guarantee a better-looking result.
Lanczos and ESRGAN Thick offer 2×, 4× and 8×. Slim and Medium offer 2× and 4×. Choosing 8× while Slim or Medium is selected switches to Thick; choosing Slim or Medium at 8× lowers the scale to 4×.
Options marked Too large exceed the app's output limits for your browser. ashUp may use a lower scale when the selected one will not fit. Check the output dimensions before starting; available memory can still limit very large images.
A model is the method ashUp uses to enlarge your image. Try these options on the same image and scale to see which result you prefer:
AI models predict plausible detail from patterns learned during training. They cannot recover missing detail with certainty. A larger model or sharper-looking texture is not always more accurate or better for your image.
The Speed and Detail dots are general guides, rather than measurements of your image or device. Actual speed depends on your hardware, browser, scale and image size.
Check both the whole image and small details. Look closely at faces, lettering, edges and repeating textures, where AI changes can be easier to spot.
Preview zoom does not change the dimensions or quality of the downloaded image.
After upscaling, open File to choose how to save the result:
The Quality control applies to JPEG and WebP. A lower setting usually makes a smaller file but can lose fine detail or introduce visible marks around edges. A higher setting does not add detail or change the output dimensions. You can change the format or quality and download again without repeating the upscale.
Download saves a file using your chosen export settings. Copy puts the result on your clipboard as PNG, regardless of the selected format, so you can paste it into another app. Copy is available only when your browser supports it. If your browser cannot save the chosen format, ashUp saves PNG instead and lets you know.
Saved images use sRGB, a common colour space for screens and the web. Camera metadata (EXIF), such as camera settings and location information, is not kept. Keep your original file if you need that information.
Your images are processed in this browser on your device. They are not uploaded or used to train the AI models. Loading the app still uses an internet connection to download app files, fonts and any AI model files you need.
Each AI model has separate downloads for its supported scales. Using Medium at 2× does not also download Medium at 4×. The model card shows the download size for the active scale, or Downloaded when those files are saved in this browser.
Once the app has been saved for offline use by your browser, you can reopen it without a connection and use Lanczos or models already marked Downloaded at the scale you need. Before going offline, load the app online and complete an upscale with the model and scale you plan to use.
Downloads are stored in this browser's storage, rather than as files in your Downloads folder. Storage may be blocked, full or cleared, especially in private browsing. If ashUp cannot save a model, it tells you; the model may need downloading again next time. Clearing site data also removes saved downloads.
Remove downloads clears the saved AI model files to free space. It does not delete your original images or files you have downloaded. You will need a connection to download a removed model again.
Why is upscaling slow? The first run may include downloading files and loading the AI model. Processing time then depends on your device, image, model and scale. If graphics acceleration is unavailable, AI can be much slower. Try Slim, a smaller scale, or Lanczos, and close other demanding apps.
Why is a scale unavailable or changed? Slim and Medium stop at 4×, and very large outputs can exceed your browser's limits. Check the active model and output dimensions. Try a lower scale or a smaller source image if the output is too large.
Why do faces, text or textures look wrong? AI can invent plausible details that do not match the original. Compare at 100% and try another model or Lanczos. Upscaling cannot reliably restore unreadable text or recover detail lost to heavy blur.
Why is a model downloading again? A different model or scale needs its own files. A download may also be needed if browser storage was cleared or the earlier files could not be saved. Check Downloaded models before relying on offline use.
What if progress stops or the upscale fails? A single tile can take time, especially on a slower device. If progress stays unchanged, cancel and try a lighter model or lower scale. Cancelling discards the current run's progress, so it must start again; a previous finished result is kept if there is one.
Why will an image not open? The file may be damaged or in a format this browser cannot read. Try opening it in another image app and exporting a PNG or JPEG copy, then add that copy to ashUp.
ESRGAN stands for Enhanced Super-Resolution Generative Adversarial Network. It is a type of AI trained to turn small images into larger images with convincing detail.
During training, one neural network learns to create larger images, while another learns to distinguish generated images from real examples. Like a student and a critic, they help each other improve. This training happens before you download the model; your images do not teach it anything.
When you upscale, ashUp uses the downloaded model to process small, overlapping sections of your image, called tiles, and joins the results together. The tile counter shows this work progressing. Everything runs on your device.
How is this different from using an AI image editor? ashUp uses neural networks specialised in enlarging an existing image. They predict higher-resolution edges and textures directly from your image, without a text prompt. Tools such as GPT Image and Nano Banana support much broader image creation and editing, guided by instructions.
ashUp's AI is designed for enlargement rather than requested changes to the scene. However, its predicted details can still be inaccurate. It cannot guarantee that a clearer-looking feature matches what was originally photographed.
The result is a prediction. The model cannot know exactly what was in the original scene. Convincing hair, fabric or lettering may be newly generated detail. Do not treat those additions as evidence of what the original photo showed.
CPU means the browser is reporting processor pressure in four levels: Low, Moderate, High and Critical. These levels describe how busy the processor is, rather than an exact percentage or the work done by ashUp alone.
Page appears when CPU readings are unavailable. It shows how responsive this page is, from Smooth to Stalling. The page can stay responsive while image processing continues, so this reading does not tell you how busy the whole device is.
GPU or CPU beside the speed reading tells you where the AI model is running. GPU means graphics acceleration is in use; CPU means it is running without that acceleration. The CPU load level can stay low while the GPU is busy.
MP/s means millions of output pixels processed per second. It measures ashUp's processing speed, not GPU load. The reading appears as tiles finish during AI processing; downloading, loading the model and saving the image are separate steps.
Compare speeds using the same image, model and scale. Higher scales create more pixels per tile, and other apps or a device slowing down to stay cool can affect the reading. To check GPU load, use your system's own tools, such as Task Manager on Windows.
This tool is built on work shared by other people. Thank you to: