The ISO Decision at 3:00 AM
You're shooting a night exterior on a single practical streetlight with your ISO at 3200. The monitor shows texture in the shadows that could be read as either noise or grain depending on what you do in the grade. You can go to ISO 6400 and gain another stop of shadow detail, or hold at 3200 and add a key light that changes the entire aesthetic of the scene.
The "just use native ISO" advice tells you nothing here. Your camera has a native ISO of 800 and a dual-native of 3200. You're already at the second native. Going to ISO 6400 means amplifying the dual-native signal by 1 stop, which has a specific and measurable effect on shadow noise. Knowing that effect before you make the decision is the difference between an informed choice and a guess that you'll discover in the grade.
This post explains how noise is actually generated in CMOS sensors, what dual-native ISO is and why it exists, when noise is cinematically acceptable, and provides a practical decision framework for pushing ISO on six commonly used cinema and mirrorless sensors. The sensor architecture data here comes from manufacturer technical documentation, Sony's dual-gain architecture patent filings, and the independently published DxOMark sensor measurement methodology.
How Noise Is Generated in a CMOS Sensor
Noise in a digital sensor comes from three primary sources: shot noise, read noise, and thermal noise.
Shot noise is unavoidable. It comes from the quantum nature of light - photons arrive at the sensor in random intervals, so even a perfectly uniform light source produces a slightly uneven signal. Shot noise scales with the square root of the signal level. If a pixel collects 400 electrons, shot noise is √400 = 20 electrons, giving a signal-to-noise ratio (SNR) of 400/20 = 20:1. Collect 10,000 electrons and the SNR improves to 10,000/100 = 100:1. Shot noise is proportionally larger in shadows (low signal) than in highlights (high signal). No technology eliminates it - it is a property of light governed by Poisson statistics.
Read noise is generated by the analog-to-digital conversion circuitry as it reads each pixel. Modern BSI CMOS sensors (used in the Sony FX3, Canon R5C, and similar cameras) have dramatically lower read noise than previous-generation sensors, often measuring below 2 electrons RMS. This is why modern cameras produce cleaner shadows at high ISO than cameras from five years ago: better ADC circuits, not just larger sensors. Read noise becomes the dominant noise source at low signal levels - that is, in underexposed shadows or at very high ISO.
Thermal noise increases with sensor temperature. Long exposures and extended recording sessions in hot environments raise sensor temperature and increase thermal noise. Cinema cameras with active cooling (ARRI ALEXA systems, Sony VENICE) perform more consistently in sustained high-ISO recording than consumer mirrorless bodies that rely on passive cooling.
When you raise ISO, you raise the sensor's analog gain before analog-to-digital conversion. This amplifies the signal but also amplifies read noise and exposes the noise floor more visibly. At a camera's native ISO, the gain circuit is optimized for the best SNR at that sensitivity level. On cameras with dual-native ISO, a second optimized gain circuit activates at the higher ISO, giving you a new SNR baseline rather than simply amplifying the first circuit's noise. Sony describes this architecture in patent filings including US Patent 10,038,866.
ISO Performance Across 6 Common Cameras
The table below provides practical guidance for push decisions on six common cameras based on their sensor architecture and published noise performance data. The "Max Recommended Push" is the point where shadow noise remains cinematically usable for most delivery formats at standard grading - not the camera's ISO ceiling. All of these cameras can go higher; the question is whether the resulting image serves the project's intended look and delivery standard.
| Camera | Base ISO | Dual-Native ISO | Max Recommended Push | Shadow Character at Max Push |
|---|---|---|---|---|
| ARRI ALEXA Mini | 800 (EI) | None | EI 3200 | Fine grain, gradable |
| ARRI ALEXA 35 | 800 (EI) | None | EI 6400 | Excellent fine grain to 6400 |
| Sony FX3 | 800 | 12800 | ISO 25600 | Acceptable at 25600, coarse above |
| Sony FX6 | 800 | 12800 | ISO 25600 | Similar to FX3, slightly cleaner |
| Canon EOS C70 | 800 | 4000 | ISO 12800 | Moderate grain at 12800 |
| BMPCC 6K Pro | 400 | None | ISO 3200 | Fine grain to 3200, harsh above |
The critical distinction between cameras with and without dual-native ISO is visible in this table: the Sony FX3 and FX6 can push to ISO 25600 with acceptable shadow character because the ISO 12800 native point resets the noise floor. The BMPCC 6K Pro, with a single base ISO of 400 and no dual-native architecture, runs into coarse noise well below what the Sony cameras can manage.
Three Production Scenarios: Different Contexts, Same Framework
Scenario 1: Documentary, Sony FX3, Available Light Interior (ISO 12800 Second Native)
A documentary DP is shooting an interview in a dimly lit hospital break room. Ambient is approximately 100 lux from overhead fluorescents. The subject will not wait for lighting to be rigged. The DP sets the FX3 to ISO 12800 in S-Log3 - the second native ISO - at T2.8 with a 180-degree shutter at 24fps.
At ISO 12800 native, the FX3's dual-gain circuit provides an effective noise floor approximately equivalent to what the standard circuit delivers at ISO 3200. ISOs between 800 and 12800 (such as ISO 3200 or 6400) are non-native: they apply digital amplification to the 800 base circuit with progressively more noise. Shooting at ISO 6400 in this situation would produce more noise than shooting at ISO 12800 native, because the non-native intermediate is worse than the second native point.
Scenario 2: Narrative Feature, ARRI ALEXA Mini, Controlled Interior (EI 800)
A controlled interior location with a practicals-and-tungsten lighting setup. The gaffer has the scene at 1,600 lux on the principal subject position. The DP shoots at EI 800 with the ALEXA Mini - the single native point - at T2.8, 1/50s.
At these settings the per-pixel electron well fills to approximately 70 to 80% capacity (based on the ALEXA Mini's typical full well capacity). Shot noise at 75% full well of a 6,000-electron well = √4,500 = 67 electrons. Read noise is approximately 3 to 4 electrons. Combined noise = √(67^2 + 3.5^2) = approximately 67.1 electrons. SNR = 4,500 / 67.1 = 67:1. This is a clean, high-quality capture. No ISO push is required or desirable. The ALEXA Mini has no dual-native architecture - pushing EI requires accepting non-native amplification at every stop above 800.
Scenario 3: Run-and-Gun News, Canon EOS C70, Mixed Available Light (ISO 4000 Second Native)
A news DP is shooting in a conference room with mixed overhead LED panels (4,500K) and daylight windows (6,500K). The principal speaker is at 600 lux; a secondary speaker in a darker corner is at 150 lux. The DP needs to cover both positions without re-lighting.
The C70's Dual Gain Output architecture provides native points at ISO 800 and ISO 4000. ISO 4000 gives approximately 2.3 stops more sensitivity than ISO 800 with a proportionally lower noise penalty than non-native amplification produces. At ISO 4000, the darker corner at 150 lux exposes the secondary speaker to approximately -2 stops below the native metering target. This is recoverable in the grade with temporal NR, because the ISO 4000 native circuit has already minimized read noise amplification at the capture stage.
How to Apply the ISO Decision Framework on Set
Step 1: Identify your camera's native ISO (and dual-native points, if applicable) before the shoot. Check the camera menu or the manufacturer's technical specification document. Confirm you're recording in a LOG profile that corresponds to the native ISO selection - recording in a picture profile with artificial noise reduction applied masks true noise while permanently destroying fine texture.
Step 2: Measure the light before adjusting ISO. Use an incident light meter at the subject position. Ask whether adding a small practical (a 60W tungsten bulb, a $40 portable LED panel) recovers a stop. Every stop of additional light delivers the same exposure improvement as a one-stop ISO increase with no noise penalty.
Step 3: If you must push ISO, use the second native ISO directly. There is no benefit to stopping partway between native points. On a Sony FX3, ISO 6400 is worse than ISO 12800 because 6400 is non-native amplification of the 800 base circuit while 12800 uses the dedicated low-light gain circuit.
Step 4: Decide your noise treatment strategy before you shoot. A pushed ISO image treated with DaVinci Resolve Temporal NR looks different from one treated with FilmConvert grain simulation. Temporal NR cleans stationary backgrounds effectively but degrades fine texture in motion (hair, fabric). Grain simulation adds organic structure but stacks on top of existing digital noise. Neither is wrong; deciding in post which approach to take means deciding without seeing the image as the colorist will see it.
Step 5: Run a camera test at your anticipated push ISO before any planned high-ISO shoot. Five minutes of testing in conditions similar to the shoot saves two hours of grade problem-solving. Record at the specific ISO, aperture, and light level you expect, and grade the test through your delivery pipeline before committing to that exposure strategy.
Pro Tips and Common Mistakes
Pro Tip: On Sony FX3 and FX6, confirm which native ISO is active before each scene. The camera displays which base ISO is in use in the ISO/Gain section of the shooting menu. Accidentally using ISO 6400 in S-Log3 (between the two native points) produces more noise than ISO 12800 native because 6400 is applying non-native gain to the 800 base circuit. The native ISO points are clean circuit optima, not just labeled steps on the scale.
Pro Tip: Chroma noise (color noise) is more visually disturbing than luminance noise at equivalent levels. Most cameras produce more chroma noise than luminance noise at high ISO because the demosaicing algorithm amplifies color channel differences in low-signal pixels. Applying Resolve's Color Noise Reduction at 10 to 20% as a starting point significantly improves the visual quality of pushed footage without destroying fine luminance texture.
Pro Tip: Pushing ISO compresses usable highlight headroom. At the Sony FX3's native ISO 12800 in S-Log3, the latitude above middle grey is reduced compared to the ISO 800 native position. If your pushed-ISO scene contains bright highlights (window light, practicals, reflections), meter carefully to avoid simultaneous shadow noise and highlight clip.
Common Mistake: Using the highest available ISO to maximize shadow detail, then applying in-camera noise reduction to clean up the result. In-camera NR is applied before recording, baking it permanently into the footage. It softens fine texture, removes subtle shadow detail, and eliminates the organic noise character that grades as grain. The image looks cleaner on the monitor but is harder to work with in post.
The fix: Record at the highest acceptable ISO with no in-camera NR. Handle noise in post with Resolve's Temporal NR or dedicated grain tools, where you have full control over the trade-off between noise reduction and texture preservation.
Common Mistake: Assuming that because a camera has 14 stops of dynamic range, high-ISO footage will always have enough shadow detail to recover. Dynamic range specs are measured at the camera's base ISO. At ISO 6400 non-native on a camera with an 800 base, you've consumed 3 stops of that dynamic range through amplification. Usable range narrows from both ends: reduced highlight headroom and a raised noise floor.
Frequently Asked Questions
What is dual-native ISO and why does it matter?
Dual-native ISO means the camera has two separate gain circuits in its sensor readout electronics, each optimized for a specific sensitivity level. At the lower native ISO, the sensor uses a standard gain circuit. At the higher native ISO, the sensor switches to a different circuit with higher gain but also a lower noise amplification ratio for that gain level. The result is a noise floor at the high native ISO that's cleaner than you'd get by electronically amplifying the low native ISO by the same number of stops. Sony describes this architecture in patent US10038866. Cameras with true dual-native ISO (Sony FX series, Canon Cinema EOS with DGO) perform measurably better at their high native point than at equivalent non-native ISOs on the same camera.
Is digital noise the same as film grain aesthetically?
No, but they can be made to resemble each other. Film grain is silver halide crystal clumping that scales with exposure: lighter areas of the frame have finer grain structure than shadows. It has temporal coherence - the grain pattern moves with the image naturally - and a warm, structured aesthetic. Digital noise is random pixel-level luminance and chroma variation with less coherent structure and a harsher appearance at high levels. At low levels on a modern sensor, fine digital noise can read as grain-like. Grain simulation tools in post (Resolve Film Grain, FilmConvert) add artificial grain structure on top of digital noise, but this requires careful calibration to avoid visually doubling the texture.
When is noise acceptable for delivery?
Acceptable noise depends on delivery format and viewing context. For 4K streaming delivery, visible shadow noise at 4K native resolution may be compressed by the streaming codec and read as compression artifact rather than aesthetic texture in typical consumer viewing conditions. For theatrical DCP, the projection environment reveals fine noise more clearly. For social media delivery at 1080p or below, downsampling provides significant noise reduction, making pushed ISO more forgiving. The practical standard: process a 5-second sample of your pushed ISO footage through the full delivery pipeline at the delivery resolution, view it on the display size the audience will use, and judge from that - not from the raw file on a grading monitor.
How does aperture interact with ISO in low light?
Aperture and ISO interact directly in terms of total light captured. Opening the aperture by 1 stop (T2.8 to T2.0) doubles the light hitting the sensor, allowing you to maintain the same exposure at 1 stop lower ISO and with reduced noise. In low-light conditions, every available stop of aperture directly trades against ISO. A T1.4 cinema lens versus a T2.8 allows 4 stops lower ISO at identical scene brightness - the difference between base ISO 800 and noise-heavy ISO 12800. Fast glass (T1.4 to T2.0) is the most effective noise reduction available on set. The Exposure Calculator models the aperture-ISO-shutter speed relationship for any shooting scenario.
Why does my camera produce more noise recording video than in stills mode?
Video recording involves continuous sensor readout at 24 to 120 times per second, generating significant heat in the sensor and ADC circuits. Higher temperature raises thermal noise. Extended video sessions cause sensor temperature to rise progressively. Cameras with active cooling (ARRI ALEXA systems, Sony VENICE) maintain consistent temperature. Consumer mirrorless cameras rely on passive cooling, which is why some overheat during extended 4K recording and why noise increases noticeably in takes beyond 20 to 30 minutes in warm environments.
Related Tools
The ISO Noise Estimator models expected shadow SNR at any ISO and aperture combination for common cinema cameras - use it to predict whether shadow areas will grade as acceptable grain or unacceptable noise before you commit to an exposure strategy. The Dynamic Range Comparison Tool provides camera-specific latitude data at different ISO settings, including the effect of non-native amplification on usable dynamic range. For complete exposure decisions including shutter speed, aperture, and ND combinations, the Exposure Calculator handles the full calculation. For the shutter speed relationships that interact with ISO in the exposure triangle, The Exposure Triangle for Cinematographers covers the complete system.
Conclusion
Native ISO is where a camera performs at its best-measured SNR. It's not where creative decisions end. The cinematographers who consistently get the most from their cameras in difficult light are the ones who know exactly what their camera does between native points, at the high native, and two stops above it. That knowledge comes from deliberate testing, not from manufacturer spec sheets read in isolation.
ISO is a gain control, not a light source. The decision to push always represents a trade: more signal against more noise, with the dual-native architecture changing where the clean floor sits but not eliminating the physics above it.
What's the highest ISO you've used on a deliberate creative choice rather than an emergency exposure fix - and what did it produce that adding a light could not have?
