Enhancing an image

Open, check the settings, enhance, export. That is the whole workflow.

1. Open

Open image… (Ctrl+O), or drag a file onto the window. JPEG, PNG, WebP and TIFF. Dragging several files at once puts them in the queue instead — see the batch guide.

The panel on the left tells you what you opened, and what you will get: dimensions, megapixels, colour profile, and the predicted output size.

Your original file is never modified. Every result is written as a new file.

2. Check the settings

Image type

Seven choices, describing the kind of image rather than the machinery:

	For
General Photo	Everyday photographs — camera, phone, landscapes, objects, animals. Litesum's fidelity-trained AI at 4×, faithful classical enlargement at 2×.
High Quality Photo	Good images that already hold detail: cropped photographs, print enlargement. The most restrained option.
Low Resolution	Small web images, heavily reduced images, old digital photographs. The most assertive of the photographic profiles.
Compressed Image	JPEGs damaged by heavy compression — social-media downloads, block boundaries, mosquito noise.
Illustration and CGI	Digital paintings, anime, drawings, rendered imagery, game assets, flat-colour artwork. Runs the illustration AI network.
Text and Graphics	Logos, screenshots, interface elements, diagrams, posters, anything with typography. Runs the illustration AI network.
Generative Photo	The generative AI photographic network, as an explicit opt-in. It invents plausible texture rather than reconstructing measured detail — see the caution it shows. Its Detail synthesis slider sets how much invention to keep: at zero you get the faithful enlargement and no AI runs; at full, everything the network adds that survives the fidelity check.

Why photographs run what they run. We measure, and the default follows the measurement. First the faithful classical enlargement beat every pretrained AI model we tested on eleven of twelve real photographs, so it became the default. Then we trained our own photographic network differently — purely to be faithful to your original, with nothing in its training that rewards inventing texture — and on the same twelve photographs it beat the classical method on all twelve, by both fidelity measures. So at 4× that network is the default, and every result is still checked against a faithful resize region by region. At 2× photographs keep the classical method: we have not trained or measured a 2× network, and we do not ship defaults we have not measured. The generative look stays available as Generative Photo if you want it and accept what it is.

These are processing profiles: seven choices running on three AI networks and the classical method, and what differs between them is which one runs and where the controls start. About shows exactly which is which. Nothing is hidden.

Choose settings automatically

On by default. Litesum Enhance measures your image — noise, JPEG blocking, edge density, how much of it is flat colour — picks an image type and sets the three strengths, then tells you why in the panel. For example:

Low resolution, compressed. A small source, so detail has to be reconstructed rather than refined.

Every value it picks is written into the same controls you can edit. Moving any slider turns automatic mode off, so your adjustments are never quietly overwritten when you open the next image.

The three controls

	What it does	When it runs
Denoise	Reduces sensor and compression noise.	Before enlargement — enlarging noise magnifies it.
Compression repair	Reduces JPEG blocking on the eight-pixel grid.	Before enlargement, and before sharpening: sharpening a compressed image first turns the block grid into a feature.
Sharpen	Restores crispness.	After enlargement, because enlargement is what softened it.

That order is applied in the pipeline rather than in the graphics driver, so the same settings give the same result on any computer.

Each group has a Reset of its own, so overdoing the sharpening does not cost you the compression repair you spent a minute getting right.

Options

Options beside Denoise and Sharpen opens the switches that shape them.

Remove colour speckle too	Coloured flecks are the most obvious kind of noise and the safest to remove, because colour carries far less of the detail your eye reads than brightness does. On by default.
Protect fine detail	Holds denoising back in busy areas, so texture is not smoothed into plastic. Turn it off if you want a heavier, flatter result. On by default.
Avoid bright outlines	Keeps sharpening off high-contrast edges. A pale rim around every letter is how sharpening ruins text. On by default.

Compression repair deals with block edges only. Ringing — the faint echo around a hard edge — and colour smearing are not repaired yet, and we would rather say so than give you a control that does nothing.

Face recovery

Photographs of people get an extra step. A specialist model rebuilds faces that are small and soft in the finished image — the kind you get in a group shot, a crowd, or an old print where a face is only a few hundred pixels across.

It is on when you open a photograph, and it decides face by face. What it decides is eligibility — whether the size rule admits the face — which is a safeguard rather than a judgement that the result will be better. That is why there is a Compare button, and why it is worth pressing. Faces Litesum Enhance can genuinely help are rebuilt at a deliberately restrained strength. Faces it would harm are left exactly as the enlargement made them, untouched — not blended, not softened.

Which faces it leaves alone, and why. The model works at 512 pixels. A face larger than that in the finished image is being shrunk, rebuilt and enlarged again, and that removes detail rather than adding it: on our own test photographs it turned deep wrinkles into smooth plastic skin and redrew a pair of wire spectacle frames thinner, with a distorted rim. So a large face gets progressively less of the effect and a face half again as large gets none. The interface says "Existing facial detail preserved" when that happens.

What it tells you. A line under the control says what was found and what was actually done — "3 faces detected · recovery applied to 1". If every face was left alone it says so; it will not tell you recovery was applied when nothing was.

Control	What it does
Face recovery	The master switch. Turn it off and it stays off for that image — through zooming, changing the output size, adjusting any other setting, applying a preset, and export. Nothing turns it back on by itself.
Strength	How much of the rebuilt face to use. Moving it takes the decision off automatic and makes it yours. Reset hands it back.
Face 1, Face 2, …	Appears when more than one face is found, so you can turn individual faces off. Each row shows a small crop of that face and where in the picture it is, so you can tell which switch is whose. A face the size rule has excluded shows its reason and cannot be switched on — nothing could make the model help it.
Compare with recovery off	Flips the switch and redoes the enhancement, so you can see both. When recovery is already off the button says Compare with recovery on, because that is what it will do.

Please look before you export a portrait. Rebuilding a face is not sharpening it. Some of the finished face is the model's work rather than your photograph's, and the higher the strength the more of it is. Moles, fine lines and skin texture are the first things to go. The restrained automatic setting exists precisely because the full strength is too much for most photographs, and there is a Compare button a few lines above this for exactly this purpose.

Opening another photograph starts again. The line and the per-face switches belong to the photograph they were measured on, so opening a new one clears them. Nothing carries a previous photograph's faces forward.

A preset that says nothing about faces changes nothing about faces. Presets saved before this feature existed have no opinion on it, and neither have the built-in ones. Applying such a preset leaves your own choice for the photograph exactly as it is — off stays off, a strength you set stays set — and where you have made no choice, the automatic behaviour stands. A preset you save yourself records what you chose, so applying it later reproduces it.

No internet needed, ever. Both face models are installed with the application. There is nothing to download the first time you use it, and the photograph never leaves your computer.

Crop

Crop turns on the crop tool. Drag on the picture to draw a rectangle; drag inside it to move it, or a corner to resize it. Everything outside dims, and the thirds are marked. The resulting pixel dimensions are shown as you drag.

Shape	Free, or one of six fixed shapes — Square (1:1), 3:2, 2:3, 4:3, 3:4 and 16:9. Changing the shape reshapes the rectangle you already have rather than clearing it.
Undo crop / Redo	Step back and forward through the crops you have tried.
Whole photograph	Stop cropping and use all of it again.

Nothing is cut from your file. The original on disk is never touched, and because Litesum keeps the whole photograph you can widen a crop again, not only tighten it — every crop is worked out from the original rather than from the crop before it.

Crop comes first in the order of work: crop, enhance, then protect. The crop tool and the brushes are never on at the same time, so turning one on turns the other off.

Two things follow from that order, and Litesum says both rather than letting you find them out:

• Changing the crop keeps what you have painted. Protection is remembered against the whole photograph, so tightening a crop keeps it on everything still inside, and widening the crop again brings back what had been hidden. Nothing is stretched to fit. What a crop does clear is the brush's Undo history, because each step in it refers to the previous crop's pixels.
• Rotating and flipping keep them too, and turn them with the picture, so protection stays on the feature it was painted on.
• Export waits for the crop to be processed. If you change the crop after enhancing, Export is unavailable until you enhance again. A different crop is a different picture, and exporting the old one would quietly hand you the wrong photograph.

Output size

The predicted output size is shown as you change it, with a warning if the job is large and a refusal if it cannot be done — and the refusal offers a size that will work instead of just complaining.

1× — original size	Improves the photograph and keeps its dimensions exactly. Denoise, compression repair, sharpening and face recovery all run; nothing is enlarged, so nothing has to be invented to fill new pixels. The right choice for a photograph that is already big enough and simply needs cleaning up.
2×	Twice the width and height. Photographs take the faithful classical enlargement here.
4×	Four times the width and height, on Litesum's fidelity-trained network for photographs.
Custom size…	Reveals one more control, so there is never more than one thing setting the size.

Under Custom size… you choose how to say it:

Percentage of the original	1% to 400%. Under 100% makes the picture smaller — and the corrections still run, so this is how you clean up a photograph and deliver it small in one step.
Exact width	You give the width in pixels; the height follows the original's shape.
Exact height	The same the other way round.

Making a photograph smaller is a real option, not a fallback. Litesum enhances first and resizes last, so denoise, compression repair, sharpening and face recovery all work on the full-size picture and only then is it reduced. The reduction uses a filter chosen by measuring halos and aliasing on the kinds of picture reduction usually spoils — photographs, fine texture and lettering.

Two honest notes about sizes that are not 1×, 2× or 4×:

• Your size and the AI's size are different things. The networks are trained at particular scales. Litesum runs the nearest suitable one and then resizes to what you asked for; it never describes your size as though a model had been trained at it.
• On those jobs the "how much did the AI contribute" note is measured before the final resize, and says so when it appears. It describes the enhancement rather than the finished file.

The built-in preset Enhance Without Resizing selects original size along with a restrained starting point for it.

3. Enhance

Enhance (Ctrl+E). Progress appears above the image. The interface stays responsive; you can cancel at any time and your image is untouched.

4. Compare

The view controls sit above the image:

• Split — a draggable divider down the frame, original on the left. Left and right arrows move it. Hold the space bar at any time to see the original full-frame, and let go to come back — the quickest way to judge whether the result is actually better. This matters most at original size, where the before and after are the same dimensions and there is no size difference to give the game away.
• Split across — the same, with the divider running across the frame and the original above. Up and down arrows move it. Use this one when the detail you are judging runs left to right: a horizon, a roofline, a line of text.
• Side by side — both images at once, sharing one zoom and one position, so a difference you see is a difference in the image rather than in how it was drawn.
• Original / Enhanced — one at a time. Escape returns to Split.
• Fit, 25 %, 50 %, 100 %, 200 %, 400 %, and the mouse wheel zooms towards the cursor. Drag to pan.

Zoomed in past the window, a thumbnail appears in the bottom-right corner with a rectangle marking the part you are looking at.

Judge at 100 % or above. At Fit, everything looks good.

What 100 % means when the result is larger

100 % is 100 % of the result. The original is drawn to the same size as the result, so a feature sits in the same place on both sides of the divider and you are comparing the same thing rather than hunting for it. At 4× that means the original is being shown four times larger than its own pixels — which is the honest comparison to make, because the question is whether the result is better than an enlargement, not whether it is bigger.

One consequence worth knowing: this view will not show you the original at its own pixel scale. Zooming always works in the result's pixels.

Changing the view never re-processes anything

Switching between Split, Side by side, Original and Enhanced, moving the divider, zooming and panning all draw the images that already exist. Nothing is enhanced again, so none of it costs time.

If you change a setting after enhancing, the result on screen is the one you already have — it says Enhanced · earlier settings and a line appears under the image saying so. Press Enhance to bring it up to date. Export stays available and saves the result you can see, named after the size that result actually is.

5. Protect what you do not want changed

Enhancement is applied to the whole photograph, and sometimes that is not what you want: a mole or a freckle you would rather keep, skin texture that has been smoothed, a sign or a label where invented detail would be wrong. The Selective enhancement controls appear under the settings once there is a result.

• Protect — paint, and that part of the picture goes back to the original photograph.
• Enhance — paint, and the enhancement comes back where you protected it.
• Brush size, Softness and Stroke strength work as they do everywhere else. A soft brush fades at its edge; a stroke at less than full strength protects partly, and painting over the same place again goes further.
• Show protected areas tints them red so you can see what you have done.
• Undo takes back one whole stroke, Redo puts it back, and Clear mask returns the whole image to the enhanced result.

Paint on the enhanced side of the comparison — the side that has the result on it. Dragging on the original side does nothing, which is deliberate: there is no mask over the original.

Three things worth knowing:

Protecting an area returns it to the original, including any face recovery there. The brush works on the finished result, not on one stage of it. If you protect a face, you are protecting it from everything the enhancement did to it.

Your protected areas stay where you put them. They are remembered against the photograph itself, so you can change the output size, press Enhance again, and the same features are still protected — not a corner of them, and not somewhere else in the frame.

What you see is what is saved. The exported file is the picture on screen, and the message after saving says so when any of it is protected.

Rotating, flipping and cropping all keep what you have painted, and turn it with the picture, so protection stays on the feature you painted it on. What they clear is the brush's Undo history, because each step in it refers to pixels of the shape the photograph was before.

6. Save your edit, so you can come back to it

Saving an edit and exporting an image are two different things, and Litesum Enhance keeps them apart on purpose.

	Saves	Writes an image	Shortcut
Save edit	your settings, your protected areas, and which way up the photograph is	no	Ctrl+S
Export…	nothing to come back to	yes — the finished picture	Ctrl+Shift+S

Use Save edit when you are not finished. It writes a small file — a few kilobytes, next to your photograph unless you choose somewhere else — holding everything you have chosen, including every area you protected. Your photograph is never written to, and neither is any image you have already exported.

Use Export… when you are. That writes the picture.

Opening a saved edit. Either open the edit file itself with Open edit… (Ctrl+Shift+O), or just open the photograph as usual: if there is a saved edit beside it, Litesum Enhance offers to pick it up and tells you when it was saved. You can always say "Open fresh" instead.

Three things worth knowing

Undo starts empty. A saved edit keeps your protected areas, not the history of how you painted them. So when you reopen one, the mask is exactly as you left it, and Undo has nothing behind it until your next stroke. Clear mask still works, and still clears everything.

You have to enhance again. The edit holds instructions, not pixels, so there is no result on screen until you press Enhance. This is deliberate: a stored result would be a second copy of the truth, and it would be wrong the moment you changed a setting.

The edit knows which photograph it belongs to. If the photograph has moved, Litesum Enhance says what it is looking for — the name and the exact dimensions — and lets you find it. If you point it at a photograph of a different size, it will open it and tell you the protected areas have not been applied, because they are stored per pixel of the original and would land somewhere you never painted.

If a save fails, the edit you saved before it is untouched. The application writes to a temporary file first and only replaces the real one once the new version is complete, so a full disk or a lost permission cannot take the previous version with it. It will say so when this happens.

7. Export

Export… (Ctrl+Shift+S). The format follows the file name you choose — .png, .jpg, .webp or .tif.

• PNG and TIFF are lossless. Use these to keep everything the model reconstructed.
• JPEG is lossy and discards some of it. The quality slider controls how much. JPEG has no transparency, so transparent areas become white.
• WebP does both, and is lossless at quality 100.

Metadata is removed by default, including GPS. Tick Keep EXIF metadata to preserve camera, date and copyright — and note that includes location. Orientation is corrected so your image is never rotated twice.

The colour profile is preserved regardless, because it changes how the image looks rather than describing it.

Presets

Eight built-in starting points for jobs people actually arrive with, and you can save your own. Built-ins are never written to disk, so an update that improves them reaches everyone and nothing you do can break the set you fall back to.

Keyboard shortcuts

The whole list. There are deliberately few, and every one is on a button you can also click.

Ctrl+O	Open image…
Ctrl+Shift+O	Open edit…
Ctrl+E	Enhance
Ctrl+S	Save edit
Ctrl+Shift+S	Export…

Ctrl+S saves your work; Ctrl+Shift+S writes a picture. If you know one other program, that is the pairing you already expect, and it is the reason Export moved: Ctrl+S on a command that flattens a photograph and keeps nothing would have taught the wrong reflex.