ilovejpg provides a JPG editor online that runs entirely in the browser, and this guide uses it as a reference point for a broader question: what can any online JPG editor really do to a photograph, and what is beyond its reach? The answer depends less on the interface than on the nature of the JPEG format itself, on the arithmetic behind each slider, and on the order in which edits are applied. Understanding those foundations is what separates an image that looks crisp and natural from one that is soft, blotchy or oddly coloured.
The sections below explain how JPEG stores an image, why repeated saves degrade it, what brightness, contrast and saturation change in plain mathematical terms, how resizing algorithms work and why enlarging an image cannot add real detail, and how to choose between JPG, PNG and WebP for the finished file. The aim is practical: each idea leads to a habit you can apply the next time you open an image.
How a JPG stores a photograph
JPEG takes its name from the Joint Photographic Experts Group, the committee behind the standard published as ITU-T Recommendation T.81 in 1992 and as ISO/IEC 10918-1 in 1994. Its design is built around a simple observation about human vision: we are very sensitive to changes in brightness, less sensitive to fine changes in colour, and largely blind to very fine, rapidly alternating detail in small areas. The format discards information along exactly those lines.
- Colour conversion. The RGB pixels are usually converted into YCbCr, which separates luma (Y, the brightness) from two chroma channels describing colour differences.
- Chroma subsampling. The colour channels are commonly stored at reduced resolution; the 4:2:0 scheme keeps one colour sample for every two-by-two group of pixels.
- Block transform. Each channel is divided into 8×8 pixel blocks, and each block is transformed with the discrete cosine transform (DCT) into 64 coefficients ranging from the average tone to the finest pattern.
- Quantisation. Each coefficient is divided by a value from a quantisation table and rounded. This is where information is permanently lost, and the quality setting controls how large those divisors are.
- Entropy coding. The rounded coefficients, many of which become zero, are compressed losslessly, typically with Huffman coding.
Baseline JPEGs store the blocks in a single top-to-bottom pass, while progressive JPEGs store several passes of increasing detail, so that a blurred version of the whole image appears early while loading. The visual result after full decoding is the same; the difference lies in how the data is ordered. Separately, most cameras add an EXIF block with information such as exposure settings, orientation and, on phones, GPS coordinates.
Because quantisation happens in every lossy save, the quality setting is not a measure of how good an image is, but of how much additional detail the next save will throw away.
What an online JPG editor can and cannot do
Any editor, online or desktop, works by decoding the JPG into a grid of pixel values, changing those values and encoding a new file. That model defines its limits. It can transform pixels it already has: move them, scale them, recolour them, combine them with new drawn elements. It cannot recover information that was never recorded or that compression has already removed. No slider can bring back a sky that was captured as pure white, read a licence plate that occupies four pixels, or turn a 640-pixel image into a sharp print-sized photograph.
| Task | Realistic in a browser editor? | Notes |
|---|---|---|
| Resize, rotate 90°, flip | Yes | Available in ilovejpg |
| Brightness, contrast, saturation, grayscale | Yes | Available in ilovejpg |
| Add a text overlay | Yes | Available in ilovejpg; text becomes part of the pixels on export |
| Convert between JPG, PNG and WebP | Yes | Available in ilovejpg, with quality and target size control |
| Crop | Yes in many editors | ilovejpg has no crop tool |
| Edit text already in the photo | Only by retouching | Requires clone or healing tools; not offered by ilovejpg |
| Remove a background or object | Depends on specialised tools | Not offered by ilovejpg |
| Recover blown highlights or crushed shadows | No | The information is not in the file |
| Upscale with genuine new detail | No | Interpolation smooths; AI tools invent plausible but unverified detail |
Being clear about these limits is useful rather than discouraging. It tells you where effort pays off. Most improvements to an everyday photo come from the right size, a sensible tonal correction and careful export settings, all of which a browser editor handles well. When a task genuinely needs retouching or object removal, it is better to recognise that early and use software designed for it.
The order of operations that protects quality
Generation loss is the cumulative damage caused by decoding and re-encoding a lossy file repeatedly. Each save requantises the DCT coefficients, and because an edited image rarely lines up perfectly with the previous rounding, a little more detail is lost and a little more artefact is added each time. Blocking in smooth areas, ringing around edges and colour blotches build up gradually. A photograph that has been forwarded through several messaging apps, each recompressing it, often shows this clearly.
Edit first, compress last
The principle is to keep the image in uncompressed form in memory while you make every change, and to encode it once at the end. ilovejpg follows this approach: rotations, resizing, tone adjustments and text are applied to the pixel data in the browser, and compression happens only when you download. The sensible sequence within that session is as follows.
- Begin with the highest-quality original available, ideally straight from the camera rather than a copy saved by another app.
- Fix orientation with rotate or flip, since every later decision depends on the image being upright.
- Resize to the final dimensions, so that tone and text decisions are judged at the size the viewer will see.
- Adjust brightness, contrast and saturation, or convert to grayscale.
- Add any text overlay, sized relative to the final dimensions.
- Choose format and quality, check the estimated file size, and export once.
Avoiding repeated saves
If you discover a mistake after exporting, return to the original rather than editing the exported JPG. Keep originals in a separate folder and treat exports as disposable outputs. Where a workflow genuinely requires several rounds of editing, an intermediate lossless format such as PNG avoids generation loss between rounds, at the cost of larger files, with a single JPG export at the very end.
Rotating a JPG by 90 degrees in a canvas-based editor re-encodes it. Specialist command-line tools can rotate JPEG data losslessly by rearranging the DCT blocks, but only when the image dimensions fit the block structure; for a single careful export the difference is rarely visible.
Brightness, contrast and saturation explained
In an ordinary 8-bit JPG each pixel has red, green and blue values from 0 to 255, giving 256 levels per channel. Every tone slider is a rule for turning old values into new ones. Knowing the rule explains both what the slider is good for and how it can go wrong. Precise formulas vary between programs, and browsers implement their own versions, so the descriptions below set out the general principles rather than one exact equation.
Brightness
Brightness moves every value up or down, either by adding a constant or by multiplying each value by a factor. Either way, the whole range of tones shifts lighter or darker. The catch is the ceiling and the floor. Values cannot exceed 255 or fall below 0, so when a pixel would go beyond those limits it is clipped. A light grey cloud at 240 and a white cloud at 250, brightened by 20, both become 255, and the difference between them is gone. Clipping is irreversible in that file, which is why brightening should stop before highlight detail flattens into plain white.
Contrast
Contrast scales the distance of each value from a midpoint, often the middle grey near 128. Increasing contrast multiplies those distances by a factor greater than one, so values below the midpoint get darker and values above it get lighter; decreasing contrast multiplies by a factor less than one, pulling everything towards grey. High contrast therefore clips both ends at once, losing shadow and highlight detail simultaneously. Moderate increases make images look crisper because edges between light and dark become more pronounced.
Saturation
Saturation is easiest to understand by comparing each pixel with its own grey equivalent, the grey of the same perceived brightness. Increasing saturation pushes the pixel's colour further away from that grey; decreasing it pulls the colour towards the grey; at zero, every pixel becomes its grey equivalent. Because the push happens in each of the red, green and blue channels, strong saturation also causes clipping, typically in one channel at a time, which flattens detail in intensely coloured areas such as flowers, neon signs and sunsets.
Grayscale
A good grayscale conversion does not simply average red, green and blue, because the eye perceives green as much brighter than blue at the same numerical value. Standard luma formulas weight the channels accordingly; the widely used Rec. 601 weighting, for example, takes roughly 30 per cent of red, 59 per cent of green and 11 per cent of blue. That is why a blue sky turns fairly dark in grayscale while green foliage stays light.
| Control | What changes, in plain terms | Typical failure when overused |
|---|---|---|
| Brightness | All values shift up or down together | Highlights clip to white or shadows to black |
| Contrast | Values are pushed away from, or towards, middle grey | Both shadows and highlights clip; skin looks harsh |
| Saturation | Each colour moves away from, or towards, its own grey | Single channels clip; colours look artificial and lose texture |
| Grayscale | Colour is replaced by a weighted brightness value | Colours of similar brightness merge into the same grey |
One further effect deserves mention. With only 256 levels per channel, stretching a narrow range of tones across a wider range leaves gaps between the levels that are actually used. In smooth gradients such as skies, these gaps appear as visible bands, a defect known as banding or posterisation. Small adjustments rarely cause it; large contrast increases on a dull image frequently do, and JPEG compression can make the bands more obvious.
Resizing algorithms: why downscaling is safe and upscaling invents nothing
Resizing means calculating a new grid of pixels from the old one. The method used to calculate each new pixel is the resampling algorithm. The main families, in increasing order of sophistication, are listed below.
- Nearest neighbour. Each new pixel copies the closest old pixel. It is fast and keeps hard edges, which suits pixel art, but produces jagged results on photographs.
- Bilinear. Each new pixel is a weighted average of the nearest two-by-two group of old pixels. Results are smooth but slightly soft.
- Bicubic. Each new pixel uses a four-by-four neighbourhood with a curve that preserves edges better. It is a common default in image software.
- Lanczos. A wider neighbourhood based on a windowed sinc function, generally sharper still, at the cost of slight ringing near hard edges.
Downscaling
Reducing an image combines many old pixels into each new pixel, so it discards information but invents none. Done well, it often makes a photo look sharper, because small amounts of noise and softness average out. The one hazard is aliasing: if the algorithm samples too few of the old pixels, fine repeating patterns such as fabric, brickwork or screen grids can turn into moiré or jagged edges. Browsers apply smoothing when drawing a scaled image onto a canvas, which reduces this risk for typical photos. For very large reductions, halving the size more than once can give cleaner results than one enormous jump in some tools.
Upscaling
Enlarging an image requires creating pixels where none existed. Interpolation fills them in by blending neighbouring values, which produces smooth transitions but no new detail; edges become soft and small text stays illegible, just larger. Doubling the dimensions of a 1,000-pixel photo gives a 2,000-pixel file containing exactly the same information. AI upscaling tools go further by generating plausible texture learned from other images, which can look convincing, but the added detail is a prediction, not a record of the scene. For evidence, documents or anything where accuracy matters, that distinction is important. ilovejpg resizes by interpolation and includes no AI features.
Use the aspect-ratio lock unless you deliberately want to distort the image. Even a small unequal change, such as 1,000 × 750 to 1,000 × 700, visibly squashes faces and circles.
Pixels, DPI and the size an image really needs
A digital image has a size in pixels; it has no physical size until it is printed or displayed. The DPI or PPI value stored in a file is only a hint about how large to print it, and changing that number alone does not alter a single pixel. What matters is how many pixels are available for the space the image will occupy.
For screens, size the image to the largest width at which it will be displayed, allowing extra for high-density displays where useful; 1,600 to 2,400 pixels on the long edge covers most full-width uses. For printing, multiply the print size in inches by the desired pixels per inch. A 6 × 4 inch photograph at 300 pixels per inch needs 1,800 × 1,200 pixels; an A4 page at 300 needs about 2,480 × 3,508; at 150 pixels per inch, which suits many documents viewed at reading distance, the A4 requirement halves in each dimension. Images larger than their destination require are not sharper when shown there, only heavier.
Choosing the output format and quality
The final decision is how to encode the edited pixels. ilovejpg exports JPG, PNG or WebP, with a quality slider for the two lossy options, a live size estimate, and an optional target size in KB that searches for the highest quality fitting under the limit. Re-encoding through the canvas removes EXIF metadata, so camera details and GPS coordinates do not travel with the exported file; if you need that information, keep it in the original.
| Format | Compression | Best for | Avoid for |
|---|---|---|---|
| JPG | Lossy, DCT-based | Photographs and continuous-tone images; universal compatibility | Screenshots, line art, text-heavy graphics, images needing transparency |
| PNG | Lossless, DEFLATE-based | Screenshots, diagrams, graphics with flat colour and sharp text | Large photographs, where files become very big |
| WebP | Lossy or lossless | Web images where smaller files matter and the destination accepts WebP | Forms or older software that require JPG |
- For JPG photos, quality between about 80 and 92 is a dependable range; above 92 file size rises steeply for little visible gain.
- If the file is still too large at around 75, reduce pixel dimensions rather than quality further.
- For text, lines and interface elements, PNG avoids the halos that JPEG quantisation creates around sharp edges.
- When a strict limit applies, combine a sensible pixel size with the target size option rather than relying on quality alone.
Frequently Asked Questions
Does editing a JPG always reduce its quality?
Changing the pixels and saving again as JPG always involves another round of lossy compression, so some detail is discarded. At high quality settings the loss is usually invisible after a single save. Problems appear when a file is edited and saved many times, which compounds the damage. Make all edits in one session from the original and export once to keep the loss to a minimum.
Can an online editor make a blurry photo sharp?
Not genuinely. Sharpening filters increase contrast along edges, which can make an image look crisper, but they cannot recover detail the camera never captured or that compression has removed. Downscaling a slightly soft photo often makes it appear sharper at its new size. AI tools can generate plausible detail, but that detail is invented rather than recovered. ilovejpg offers no sharpening or AI enhancement.
Why do my highlights look flat after increasing brightness?
Pixel values in an 8-bit image cannot exceed 255. When brightness pushes light areas past that ceiling, they are clipped to pure white, and different shades that once held texture become identical. The same happens to shadows at zero when darkening or raising contrast. Use smaller adjustments and watch the brightest areas on the preview so that clouds, skin and white fabric keep their detail.
Is it safe to enlarge a small JPG for printing?
You can enlarge it, but you will not gain real detail. Interpolation blends neighbouring pixels to fill the new grid, so the printed image looks smooth and soft rather than sharp. For a good print, aim for around 300 pixels per inch at the printed size, and 150 for documents viewed at arm's length. If the original falls well short, print it smaller instead.
Should I use JPG, PNG or WebP for an edited image?
Use JPG for photographs that need to be accepted everywhere, PNG for screenshots, diagrams and graphics with sharp text or flat colour, and WebP for web images where the destination supports it and small files matter. JPG and WebP let you set a quality level, while PNG is lossless. Choose by content type first, then fine-tune quality or pixel size to meet any file size limit.
Does the editor keep my photo's EXIF data?
No. When ilovejpg re-encodes an image through the HTML5 Canvas, the EXIF block is not carried over, so camera model, exposure settings, date and any GPS coordinates are removed from the exported file. That is useful for privacy when sharing photos publicly. If you need to keep the metadata, store the original file separately, since the exported copy cannot restore it.
Conclusion
An online JPG editor is best understood as a precise tool for transforming pixels that already exist. It can resize, rotate, flip, adjust tone and colour, add text and re-encode in a better format, and in a browser-based tool such as ilovejpg it does all of this without uploading your files. What it cannot do is restore information that the camera never recorded or that compression and clipping have discarded, and no amount of upscaling turns interpolated pixels into genuine detail.
Quality is protected less by any particular slider than by discipline. Start from the original, correct orientation, resize to the final dimensions, make moderate tonal adjustments while watching highlights and shadows, add text last, and encode once in the format that suits the content. Keep originals untouched and treat every export as disposable. Those habits, grounded in how JPEG's blocks, quantisation and 8-bit values actually behave, produce images that look clean on every screen and print as well as their pixels allow.