Lift a Photograph Shot in Poor Light
Two photographs can look equally dark on your screen and need opposite treatment. One was recorded cleanly and simply printed too dark, so everything you want is sitting in the file waiting to be pulled up. The other was recorded at the sensor's limit, where the camera was already amplifying nearly nothing, and lifting it exposes the noise instead of the scene. Telling them apart takes ten seconds and decides what you can expect.
Use this without the search next time. Prathom Workbench puts Prathom's tools in your toolbar.
Add to Chrome — freeDrop an image here, or click to browse
Up to 25 MB. Uploaded temporarily, deleted after processing.
What it does
- Shadows opened without flattening the scene
- Sensor noise controlled as the exposure lifts
- Color of the original light sources preserved
- Night scenes kept looking like night
How to use Improve Low Light Photos
- 1
Check the shadows at full size first
Zoom to one hundred percent on a dark area. Smooth dark means the data is there. Colored speckle means the sensor was already at its limit.
- 2
Run the lift
The picture is repainted at a moderate denoise, opening the shadows while holding the color of the lamps and the character of the light.
- 3
Judge whether it still reads as night
A good result is a readable dark scene. A bad one is a gray daytime picture of the same subject with the atmosphere removed.
How it works
Your image is scaled to about one megapixel and repainted at a moderate denoise with a fresh seed each run. The instruction asks for the shadows to be opened so the subject is readable, for the noise the underexposure was hiding to be controlled as that happens, and for the color of the light sources to be held as they were. The negative prompt argues against the three ways this goes wrong: washed out flat shadows, color blotches in the dark areas, and a daylight look.
The two halves of the instruction fight each other on purpose. Lifting exposure and suppressing noise pull in opposite directions, and the point of running them in one pass is that the model can trade them off locally rather than applying one and then the other.
Two problems that look identical
The distinction matters more than any setting here.
An underexposed photograph was recorded correctly and rendered dark. The camera collected plenty of light, the sensor was working comfortably, and the scene is all there in the file at low values. Lifting it is close to free: the detail appears because it was always present, and the noise floor stays where it was.
A light-starved photograph is different. There was not enough light in the room, so the camera raised its gain to compensate, amplifying both the faint signal and the sensor's own electrical noise together. The file already contains everything the sensor could offer, and much of what is in the shadows is not scene, it is noise. Lifting reveals that noise, and cleaning the noise takes the remaining detail with it.
Both produce a dark thumbnail. Only one of them has a good outcome available, and the check is to zoom into a shadow and see whether it is smooth or speckled.
Why night should stay night
There is a temptation in every brightening tool to go too far, and it is worth resisting on purpose.
A photograph taken at dusk, in a restaurant, or under street lamps has a character that comes from the light being uneven and colored. Pools of warmth, deep surrounding dark, faces lit from one side. That contrast is the picture.
Flatten it and you get an image where everything is equally visible and nothing is interesting — a gray record of a room rather than a photograph of an evening. This workflow is instructed to keep the light sources colored and the falloff intact, so the aim is a scene you can read rather than a scene that has been leveled.
If the output looks like it was taken at noon, that is a failure even though it is brighter.
The fix that beats every tool
Almost all of this is decided before the shutter.
Bracing the camera against a wall or a table lets you use a slower shutter at a much lower gain, which is the single biggest improvement available. Turning on the phone's night mode, if the scene is still, collects several frames and merges them, which genuinely adds light rather than amplifying what is there. Moving one step so a lamp is behind you instead of behind the subject changes the exposure problem entirely.
None of that helps with a photograph you already have, which is what this page is for. It helps enormously with the next one.
Publication gate
This page ships once the workflow has been run against a clean underexposed frame, a high ISO night street, a mixed scene with warm lamps and deep shadow, and a correctly exposed photograph, confirming in the last case that the pass leaves a good exposure alone.
Examples
Underexposed indoor portrait
The good case. The scene was recorded properly and merely rendered dark, so lifting it reveals real detail rather than amplifying noise.
Handheld night street
Partial. Noise control does most of the work here, and detail in the darkest corners is reconstructed rather than revealed.
Frequently asked questions
How do I tell which kind of dark photo I have?
Open it at full size and look into a shadow. If it is smooth and simply dim, the exposure was low but the recording was clean, and lifting will work well. If it is full of colored grain, the sensor was already amplifying hard and there is little underneath, so lifting mostly reveals noise.
Why does my phone's night mode look better than this?
Because it is not doing the same thing. Night mode captures a burst of frames over a second or more and merges them, which genuinely collects more light and lowers noise by averaging. That happens at capture time. Once there is a single dark file, that opportunity has passed and nothing recreates it.
Will it make the picture look like daylight?
It should not, and that is deliberate. The instruction holds the color of the original light sources, so lamps stay warm and a street stays a street at night. A lifted photograph that reads as flat daylight has lost the reason it was taken, which is usually worse than being slightly dark.
Can it fix a picture that is nearly black?
Rarely. When the shadows are clipped, all three channels have already been recorded as zero and there is no difference between one dark thing and another. What comes back there is generated from context. Treat a very dark frame as a reconstruction rather than a recovery.
Is the AI Improve Low Light Photos tool free, and do I need an account?
No sign-up. The improve low light photos workflow takes an image up to 25 MB, runs it on the GPU host and hands back a file you can download directly. Nothing is charged and nothing is stamped onto the output. Because no account exists, nothing keeps your history either, so save the result within the thirty minutes the job is retained.