Why Do High-ISO Photos Look Grainy? Light, Sensors and Noise Explained

When photographing someone moving indoors, you use a faster shutter speed and raise ISO. The photo stays bright, but its shadows look grainy. A night scene shot on a tripod, meanwhile, can look clean at low ISO.Blaming that difference simply on high ISO misses the real cause.

This article is for photographers who already use shutter speed, aperture and ISO. It explains the sensor behavior behind those settings; no circuit knowledge is needed. Let’s start with the photos, then look inside the camera.

Keep two things separate: how much light you capture, and how brightly you render the resulting signal.

Why does a high-ISO photo look grainier at the same brightness?

Figure 1The same brightness does not mean the same amount of captured light

Imagine photographing the same window at night with the same aperture and lighting. Make the shutter speed two stops faster and raise ISO two stops: the output brightness will be roughly the same.

Capture more light

f/2.8 · 1/125 s · ISO 800

Schematic illustration of grain at a nighttime windowTwo illustrations of the same window. Grain on the dark wall and background shows that apparent brightness and graininess are separate. These are neither photographs nor camera image-quality measurements. Same output brightness

Light received: 1× baseline
Lower relative graininess

Offset a faster shutter with ISO

f/2.8 · 1/500 s · ISO 3200

Schematic illustration of grain at a nighttime windowTwo illustrations of the same window. Grain on the dark wall and background shows that apparent brightness and graininess are separate. These are neither photographs nor camera image-quality measurements. Same output brightness

Light received: ¼ baseline
Same brightness, more visible grain

Brightening the photo with ISO cannot replace the light missed during the shorter exposure.

What if you raise only ISO?

Keep shutter speed, aperture and lighting fixed.

Schematic illustration of grain at a nighttime windowTwo illustrations of the same window. Grain on the dark wall and background shows that apparent brightness and graininess are separate. These are neither photographs nor camera image-quality measurements. Display before brightening

1/125 s · f/2.8 · ISO 800
Captured light stays the same

Schematic illustration of grain at a nighttime windowTwo illustrations of the same window. Grain on the dark wall and background shows that apparent brightness and graininess are separate. These are neither photographs nor camera image-quality measurements. Display brightened by ISO alone

1/125 s · f/2.8 · ISO 3200
Captured light stays the same

What changes is signal handling and output brightness. Compare at the same final brightness, and the proportion of light-derived noise stays the same. On a real camera, read noise and saturation matter too.

These are schematic illustrations with exaggerated grain. They are not photos or simulations of a real camera. This simplified, non-saturating example treats a two-stop ISO change as 4× signal gain. JPEG tone curves and noise reduction are not reproduced. The top pair compares equal output brightness; the ISO-only pair shows a change in brightness.

Changing from 1/125 s to 1/500 s gives the sensor one-quarter as much light at the same aperture and lighting. Raising ISO from 800 to 3200 can brighten the output, but it cannot restore the missing light.

So don’t take away the rule that higher ISO always means more noise.Changing only ISO and reducing exposure while changing ISO are different comparisons. Autoexposure and Auto ISO can make both changes happen together.

Why does less light mean more grain?

Pixels use electrons produced by light as their signal. But the electron count is never exactly identical from one exposure to the next. Even a uniform wall varies slightly in brightness, which can look like grain. This unavoidable variation in light is called photon shot noise.

Figure 2With less light, small fluctuations stand out more

Even when you photograph a uniform wall, pixels do not receive exactly equal amounts of light. Slight differences in the number of photoelectrons appear as brightness grain.

Less light: 100 electrons on averageBrightness varies even across a uniform areaFor pixels averaging 100 electrons, theoretical standard deviation is 10 electrons, or 10%. The three examples show the mean minus one standard deviation, the mean, and the mean plus one standard deviation. They are not measurements of three actual pixels. Three enlarged pixels 90100110 Relative fluctuation: 10/100 = 10%

A 10-electron fluctuation is 10% of the signal.

Four times the light: 400 electrons on averageBrightness varies even across a uniform areaFor pixels averaging 400 electrons, theoretical standard deviation is 20 electrons, or 5%. The three examples show the mean minus one standard deviation, the mean, and the mean plus one standard deviation. They are not measurements of three actual pixels. Three enlarged pixels 380400420 Relative fluctuation: 20/400 = 5%

Even a 20-electron fluctuation is only 5% of the signal.

Capture more light and the absolute variation grows, but it becomes smaller relative to the signal.

The numbers behind shot noise and SNR

In an ideal Poisson model, if the mean electron count is N, the standard deviation of photon shot noise is √N: 10 electrons for 100, and 20 for 400.

Signal-to-noise ratio is called SNR. With shot noise alone, SNR = N/√N = √N. Capturing four times the light doubles SNR from 10 to 20.

The three pixels in the diagram illustrate mean ±1σ; they are neither measurements nor random samples. Read noise, dark current and fixed-pattern effects are excluded from this comparison.

The statistical variation in light is called photon shot noise. Electron counts and gray levels are assumed for explanation. The diagram is not meant to estimate an actual standard deviation from three values.

In practice, capture more light where you can. Open the aperture, lengthen the exposure, or change the lighting or camera position. But lengthening the exposure until a moving person blurs defeats the purpose.

So far, the diagrams have isolated noise caused by light. Real images also contain noise from the readout circuitry described below, and long exposures add dark-current effects.

What changes inside the sensor when you change ISO?

A sensor converts light from the lens into electrical signals that can be recorded as an image. Light produces electrons in tiny pixels; their quantity is read as a voltage and ultimately converted into numbers.

Figure 3Inside the camera, light becomes an electrical signal

A sensor is an array of small pixels that convert light into electrical signals. They collect electrons produced by light, then read their quantity as a voltage.

① From lens to sensorSensor pixels receive light from the lensSide-view schematic of a camera. A blue image sensor behind the lens receives incoming light. The enlarged view below shows small pixels across the sensor surface. Other parts, including the shutter, are omitted. LensSensor Array of tiny pixels

Yellow represents light and blue the sensor. Structure and scale are schematic.

② Inside one pixelStore photoelectrons, then move them for readoutYellow wavy lines are photons; circles with minus signs are electrons. The photodiode on the left stores electrons, then a transfer gate opens after exposure to move them to the readout node on the right. Electron transfer lowers that node’s absolute voltage. Conceptual diagram of a representative 4T CMOS pixel. LightRead out after exposure Transfer gate PhotodiodeReadout node Voltage before transfer Voltage drops

A minus sign inside a circle marks an electron. Clear the checkbox to return to the pre-transfer view. Electron counts are illustrative, not actual counts.

③ Amplify the voltage signalThe same electron count can produce a larger voltage signalAssume 100 electrons produce a 1 mV signal. Amplifying the magnitude of the voltage difference from its reference by 4× makes it 4 mV. The number of signal electrons does not increase. The diagram does not show the FD’s absolute voltage rising. No extra electrons 1 mV4 mV ×4 Original signalAmplified signal Signal voltage magnitude

On some cameras, the ISO setting changes this amplification, among other things. The diagram assumes the same 100 electrons and a 1 mV → 4 mV signal.

④ Convert to numbers for RAWTurn voltage into discrete numbers for RAW recordingAn assumed 3-bit ADC maps 0 to less than 8 mV into codes 0–7 in 1 mV steps. A 4 mV input gives code 4. These are not actual bit-depth or RAW specifications. Record voltage as a number 047 4 mV8 CodeInput voltage (mV)

This voltage-to-number conversion is called ADC. The 3-bit diagram is for explanation, not a camera specification. RAW recording and processing vary by camera.

ISO settings change how the captured signal is read and mapped to output, not how much light was captured.

Circuit terms and omissions

In ②, the light-receiving area is the photodiode (PD), the readout node is FD/SN, and the control between them is the transfer gate (TG). The schematic uses a representative 4T CMOS pixel and omits reset, source follower and row selection.

Electron transfer lowers the FD’s absolute voltage. Correlated double sampling (CDS) usually measures the difference between voltages after reset and after transfer. In ③, that difference is shown as a positive magnitude; the FD voltage is not shown rising. The order of CDS and ADC depends on the design.

ISO control can involve conversion gain, analog gain, digital gain after ADC, and RAW metadata. The same amplification stage does not operate in every camera, and an ISO number is not the gain factor of one circuit.

These original diagrams are based on a representative CMOS layout. Photons and electrons use different symbols. The transfer diagram assumes exposure has ended and the readout node has been reset. Not every photon is detected.

Raising ISO does not multiply the electrons produced by the same amount of light.Amplification, readout settings and the mapping to output brightness can change. Implementation varies by camera, so ISO 3200 does not imply that any one particular circuit must be active.

For a deeper look at pixels and circuitry, see How Noise Arises in CMOS Image Sensors. Here, we’ll focus on what matters for shooting decisions.

Is shooting at low ISO and brightening RAW later the same?

You might ask: if the light captured is the same, why not shoot a dark image at low ISO and lift it later? The answer depends on the camera and ISO range.

As the camera converts light into an electrical signal and reads it out, the circuitry adds small fluctuations. This is read noise. If the signal can be amplified before some of that noise is added, the same light may produce a cleaner shadow recording.

Figure 4Raising ISO in camera can differ from brightening in post

Compare RAW captures made with the same light at the same final brightness. Differences can arise because the readout circuit also adds small fluctuations. The traces isolate noise added after amplification in a schematic example.

Shoot low ISO; lift in postPost-gain noise at equal final brightnessSchematic example with the same signal normalized to final brightness 100. At 1× analog gain, post-gain noise with standard deviation equivalent to 8 becomes 8 when referred to the input. Only this trace omits photon shot noise and pre-gain noise. Match final brightness 85100115 Repeat (schematic)Signal (relative)

Brightening after recording also enlarges noise added during readout.

Amplify the signal in camera firstPost-gain noise at equal final brightnessSchematic example with the same signal normalized to final brightness 100. At 4× analog gain, post-gain noise with standard deviation equivalent to 8 becomes 2 when referred to the input. Only this trace omits photon shot noise and pre-gain noise. Match final brightness 85100115 Repeat (schematic)Signal (relative)

The signal is larger relative to noise added later, so after matching brightness that noise may have less effect.

For some cameras and ISO ranges, higher ISO can produce cleaner shadow readout with the same captured light. It cannot remove grain caused by too little light.

Numerical example: noise before and after gain

The traces show only noise added after gain in a schematic example. They exclude photon shot noise and pre-gain noise. The plotted points are neither measurements nor rigorous statistical samples.

In a simplified model that includes those sources, assume a mean signal of 100 electrons, 2 electrons of pre-gain noise, and post-gain noise equivalent to 8 input electrons at gain 1. At gain 1, total variance is 168 electrons² and SNR ≈ 7.7; at gain 4, variance is 108 electrons² and SNR ≈ 9.6.

SNR = N/√[N + r_pre² + (r_post/g)²]. Assuming independent noise sources, we add variances, not standard deviations. If post-gain noise is negligible, both cases have variance 104 electrons² and SNR ≈ 9.8, so the difference shrinks.

These are not measurements of a real camera. The model omits conversion-gain switching, amplifier noise, dark current, quantization and saturation, among other factors.

The diagram does not claim that raising ISO always helps. It assumes a particular location for analog amplification and a contribution from later noise. Existing fluctuations in light are amplified along with the signal.

Conversely, on some cameras and within some ISO ranges, RAW files made at the same exposure show little difference in shadow noise after matching final brightness. This is called ISO invariance. It does not mean every ISO has identical image quality or that capturing too little light is harmless.

A dark low-ISO RAW file also makes exposure harder to judge in the field. JPEG tone and noise-reduction processing can change, so a RAW comparison does not automatically apply to JPEG.

When lifting shadows at high ISO, check highlights too

Making the signal larger cannot extend bright-end recording without limit. Both the number of electrons a sensor can store and the capacity of its readout and digitization have limits. Once a limit is reached, distinctions between brightness levels are lost.

Higher ISO can reduce highlight headroom, but shadow read noise also changes. You cannot assume that every one-stop ISO increase always costs exactly one stop of dynamic range. Some cameras change behavior at a gain switch.

Look at both shadow grain and the highlights you want to keep. JPEG histograms and clipping warnings do not always match RAW saturation. For more, see How HDR Relates to Dynamic Range.

Technical note: high ISO and highlight headroom

A pixel running out of room for electrons is different from the circuitry or ADC after amplification reaching its output limit first. Here is a simplified model with a fixed output ceiling of 1.

Gain ×1Amplification and clipping at a fixed output limitThe horizontal axis is relative input signal from 0 to 1; the vertical axis is relative recorded output from 0 to 1. Assume gain 1 or 4 and an output limit of 1. Beyond the limit, different inputs produce the same output. Gain ×1 · same limit 011 Recorded outputLimitInput signal (relative)
Gain ×4Amplification and clipping at a fixed output limitThe horizontal axis is relative input signal from 0 to 1; the vertical axis is relative recorded output from 0 to 1. Assume gain 1 or 4 and an output limit of 1. Beyond the limit, different inputs produce the same output. Gain ×4 · same limit 011 Recorded outputLimitInput signal (relative)

At 4× gain, input ¼ reaches the same output ceiling. These are not measured ISO curves. Conversion gain, RAW recording and digital processing can change the behavior. JPEG clipping indications and RAW saturation do not necessarily agree.

Technical note: what does dual gain change?

Some sensors switch their conversion gain: the rate at which electrons become voltage. The following is one example that changes capacitance at the readout node.

Low conversion gainSwitch capacitance to change the voltage from the same electronsSimplified circuit with 2.5 fF base capacitance at the FD readout node and an extra 7.5 fF connected by a switch. Connected: 10 fF total and 1.60 mV change. Disconnected: 2.5 fF and 6.41 mV. Assumes 100 electrons and an ideal capacitor model. Connect added capacitance FD Base capacitance2.5 fF Connected Added7.5 fF 100 e⁻ · C = 10 fF |ΔV|＝1.60 mV
High conversion gainSwitch capacitance to change the voltage from the same electronsSimplified circuit with 2.5 fF base capacitance at the FD readout node and an extra 7.5 fF connected by a switch. Connected: 10 fF total and 1.60 mV change. Disconnected: 2.5 fF and 6.41 mV. Assumes 100 electrons and an ideal capacitor model. Disconnect added capacitance FD Base capacitance2.5 fF Disconnected Added7.5 fF 100 e⁻ · C = 2.5 fF |ΔV|＝6.41 mV

In an ideal capacitor model, |ΔV| = Nq/C. For 100 electrons and capacitances of 10 fF and 2.5 fF, the voltage changes are 1.60 mV and 6.41 mV. Neither the number of electrons nor the incoming light increases. These are assumed values, not camera measurements.

The capacitance shown belongs to readout node FD/SN, not the photodiode’s storage capacity. The bottom of the circuit is a reference point in a small-signal model. Parasitic capacitance and source-follower gain also affect real conversion gain.

Dual Conversion Gain, Dual Native ISO and Canon Dual Gain Output (DGO) are not the same method. Switching points and behavior vary by camera, stills versus video, and Log settings.

How should you choose ISO when shooting?

Rather than aiming for low ISO itself, keep the shutter speed and aperture you need, and capture as much light as practical. Then use the ISO setting the shot requires.

Figure 5Prioritize getting the shot over keeping ISO low

When in doubt, this order helps you separate captured light from signal handling.

First, decide what motion must be frozen

Set shutter speed for subject and camera movement. Don’t sacrifice a needed speed just to lower ISO.

Next, choose the aperture you need

Choose an f-number for blur, depth of field and lens rendering. A wider aperture can collect more light if it suits the shot.

See whether you can capture more light

Change lighting or camera position. For a still subject, use a tripod to lengthen exposure. If that is impractical, move on.

Use the ISO needed to make the shot

Auto ISO can help. Set the output brightness you want and check your camera’s normal ISO range.

Check clipping, noise and blur

If blurry, revisit ①; if light-starved, revisit ③; if highlights clip, reconsider exposure and ISO. Don’t decide by one number alone.

In aperture- or shutter-priority mode, start with the setting you care about most. Flash, video shutter angle, and Log or EI settings bring additional constraints.

This is a guide for general photography; shooting mode and subject can change the order. High ISO alone cannot make up for light you did not capture.
Shooting situationWhat to prioritize
Moving people or sportsSecure the shutter speed you need first. A sharp high-ISO photo is often better than a blurred low-ISO one.
Still night scene or productA tripod or extra lighting may let you use a normal low ISO. For long exposures, watch for dark current and subject movement.
High-contrast sceneCheck highlights as well as shadows before raising ISO. RAW saturation and JPEG clipping indications may differ.

For example, if you need 1/500 s to freeze a child indoors, keep that speed. If opening the aperture still leaves too little light, raise ISO. Blur and noise are different problems; lowering ISO is not always the answer.

Image stabilization reduces camera movement; it cannot freeze a moving subject. For more on shutter types, see Electronic, Mechanical and Electronic Front-Curtain Shutters Explained.

Common questions about ISO and noise

Does raising ISO always increase noise?

No. Distinguish reducing exposure while raising ISO from changing ISO alone. In RAW comparisons at the same exposure and final brightness, the higher-ISO image can sometimes have less read noise.

Does the lowest ISO always give the best image quality?

The lowest displayed ISO may be an extended setting, and normal base sensitivity depends on camera and mode. Blur and underexposure can also harm a photo more than the ISO number itself. Judge by the normal ISO range and shooting conditions.

Are larger pixels or sensors better at high ISO?

At the same illumination, exposure time and efficiency of converting light into electrons, a larger pixel can generally collect more light per pixel. Whole-image comparisons also depend on sensor area, pixel count, settings for the same framing, display size, downsampling and readout performance. Size alone cannot rank cameras.

Does switching to Dual Native ISO always make the image cleaner?

It depends on camera, shooting mode and Log settings. Shadow readout performance may change at the same exposure, but photon shot noise does not disappear. Check the official manual for that mode’s base sensitivity and limits.

How can I test my camera’s ISO invariance?

Photograph a still subject on a tripod in RAW with lighting, shutter speed and aperture fixed. Change only ISO, match brightness in processing, and compare shadows. Keep white balance, noise reduction, sharpening and display size consistent; exclude saturated areas.

Takeaway: separate image brightness from captured light

Raising ISO can brighten a photo, but it does not increase the light reaching the sensor. With less light, fluctuations make up a larger share of the signal. Depending on readout, however, a higher ISO may record cleaner shadows.

When shooting: prevent blur → choose the aperture you need → gather light → adjust ISO → check clipping and noise.This sequence gives you more useful choices than simply avoiding high ISO.

Disclaimer: This article is a general explanation to help photographers understand the principles. ISO control, RAW recording, base sensitivity and gain switching depend on the camera, firmware and shooting mode. For specific settings, follow your camera’s official manual.

スポンサーリンク

Comments

Copied title and URL