What Auto Enhance actually changes
Most photos that look disappointing are not badly composed — they are badly exposed, slightly cool or warm, and a little flat. Cameras meter for an average scene, so a bright window behind your subject produces a dark face, an indoor shot under warm bulbs comes out orange, and an overcast day yields something grey and lifeless. None of that is a failure of the photograph; it is a gap between what the sensor recorded and what your eye saw.
Auto Enhance closes that gap in three passes. It corrects exposure by finding where the image's tones actually sit and redistributing them so shadows have depth and highlights have detail. It corrects white balance by estimating what should be neutral in the scene and removing the colour cast that is tinting everything. And it lifts contrast and saturation by an amount derived from the image rather than a fixed number, which is what stops the result looking artificially punchy.
That is the whole of the free stage, and for most photos it is all that is needed. The two model-driven stages — denoise and face enhancement — are separate, optional, and priced individually, because they help dramatically on the right image and actively hurt on the wrong one.
Why this is not a filter
A filter is a fixed transformation. It adds the same warmth, the same contrast curve, the same fade to whatever you give it. That is why a preset that makes one photo look cinematic makes the next one look muddy: the adjustment was designed for a particular kind of source, and your image may not be that kind.
Automatic correction inverts the order of operations. It measures your image first — the distribution of brightness across the frame, whether the neutral tones skew warm or cool, how much of the histogram is bunched at either end — and then computes an adjustment specific to those measurements. Two photos from the same session get different corrections if they need different corrections.
The practical consequence is that Auto Enhance is a good default for a batch of mixed images, where a single preset would inevitably suit some and spoil others. It is a weaker choice when you want a look rather than a correction. If you are after a consistent stylistic treatment across a set, the filters tool applies a deliberate grade; if you want each photo to simply look like what you saw, this is the one.
Denoise: what it fixes and what it costs you
Image noise is the speckled, grainy texture that appears when a sensor is pushed — high ISO, dim light, a small phone sensor at night. It sits most visibly in flat dark areas, where there is no detail to distract from it, and it becomes more obvious the moment you brighten a photo, which is why an underexposed image often looks worse after correction than before.
The denoise stage is a model trained to tell noise apart from texture and remove one without the other. On a genuinely noisy photo the effect is substantial: skies clean up, shadows go smooth, and the image stops looking like it was shot in a hurry.
The cost is that no denoiser is perfect at that distinction. Fine, low-contrast detail — skin pores, fabric weave, distant foliage, the grain of wood — looks statistically a lot like noise, and some of it will go. On a noisy image that trade is clearly worth it. On a clean, well-lit image there is no noise to remove, so the only thing the pass can do is soften real detail. It costs 1 credit either way, which makes turning it on reflexively a small, repeated waste.
The rule that serves best: look at the shadows at 100% zoom. If they are speckled, denoise. If they are smooth, do not.
Face enhancement on everyday photos
Faces are the part of an image people judge hardest. We read facial detail with far more sensitivity than we read anything else in a frame, so a portrait where the eyes are slightly soft registers as a bad photo even when the rest is technically fine. General correction cannot address this, because exposure and white balance do not affect sharpness.
The face enhancement stage runs a dedicated restoration model over regions detected as faces, rebuilding structure in eyes, lashes and lips. On a group shot where each face occupies a small part of the frame, or a photo taken from across a room, the difference is obvious and welcome.
Its failure mode is over-application. On a face that is already sharp, the model has nothing to recover, so what it does instead is regularise skin texture — removing pores and fine lines and producing the smooth, slightly synthetic look associated with heavy retouching. Some people want that; many do not, and nobody wants it applied without having asked. It is off by default, costs 1 credit, and is worth switching on when the faces are the reason the photo is not working.
Choosing between enhance, upscale and restore
These three tools are easy to confuse because all three make a photo look better, but they solve different problems and cost different amounts.
Auto Enhance fixes how a photo is lit and coloured. Reach for it when the composition and sharpness are fine but the image is dark, flat, or has a colour cast. It is the cheapest of the three — free unless you enable an optional stage — and the right first thing to try.
Upscale fixes how many pixels a photo has. Reach for it when the image looks correct but is too small for what you need: a print, a large screen, a crop from a wider shot. It costs 2 credits at 2× and 4 at 4×, and it does nothing for exposure.
Restore fixes damage. It targets old, faded, scratched and low-resolution scans, rebuilding faces and detail at 2× in a single pass, and costs 3 credits. If you are working with family photographs or archival material, it is the one built for that job.
They also combine sensibly. A common sequence for an old snapshot is restore first to rebuild what is damaged, then enhance to correct the colour of the result. For a modern photo that is merely small and dull, enhance then upscale is the cheaper order — correcting a smaller image costs the same and gives the upscaler a cleaner source to work from.
How your image is handled
Auto Enhance runs on our servers rather than in your browser. The correction models are far too large to ship to a web page, so the image is uploaded over HTTPS, processed, and returned to you as a PNG. Every AI tool on OpusImg is labelled server-processed for exactly this reason, and the label is a real distinction rather than boilerplate: the client-side tools genuinely never transmit your file.
That makes the choice a per-image one rather than a per-account one. For an ordinary photo the upload is unremarkable. For something confidential — an unpublished shoot, a document, a client's material under embargo — the local tools remain available and send nothing anywhere. Enhanced images are not used to train models and are not shared.
On cost: the correction that most images need is free, permanently and on every plan, including without an account. Credits are only consumed by the two optional stages, at 1 each, and the exact figure appears before you start. Failed jobs are not charged.