Image Noise Generator
Add film grain or sensor-style noise to a photo, or make a plain noise texture. Gaussian or uniform noise at a standard deviation you set, mono or colour, with a seed so the same settings always give the same grain.
How to Use
- Choose Add grain to put noise on a photo, or Noise texture to make a plain noise image of any size. A sample photo is shown until you drop your own.
- Set Amount σ, the standard deviation in 0–255 levels: 4–8 is subtle, 12–20 is clearly visible film grain, 40+ is a heavy texture.
- Pick Gaussian (bell-shaped, like real grain and sensor noise) or Uniform (flat, good for hiding banding), and Mono (same offset on R, G and B, like film) or Colour (each channel separate, like a phone in the dark).
- Set Grain size for coarser specks, and change the Seed for a different pattern. The same seed and settings always give exactly the same result.
- Check the 100% crop under the preview, then save in any format; PNG keeps fine grain best. Show Work gives the random draws and the measured noise for your own image.
Centre crop, 100% (shown 2×)
Worked Example
Mono Gaussian grain at σ = 12 on a mid-grey pixel of 128. The generator draws two uniform numbers, say u₁ = 0.3 and u₂ = 0.8. Box–Muller gives z = √(−2 ln 0.3) × cos(2π × 0.8) = 1.55176 × 0.30902 = 0.47952, so the offset is 12 × 0.47952 = 5.75 and the pixel becomes 128 + 5.75 = 133.75 → 134, on all three channels.
Near white, noise is clipped. A pixel at 250 can only go up 5 levels; with σ = 12 about 34% of draws would push it past 255, so they stop at 255 and the measured noise there is smaller than the σ you set. That is why the Measured σ readout can come out below the Set σ on bright or dark photos. Uniform noise with the same σ spreads evenly over ±12√3 = ±20.8.
The common mistake: saving fine grain as a low-quality JPEG. JPEG compression throws away exactly the fine, pixel-to-pixel detail that grain is made of, so the grain turns into blotches and the file gets larger. Save as PNG, or as JPEG at quality 90 or above.
Show Work
Formulas
Grain, Sensor Noise and the Bell Curve
Film grain is physical: the image in a photographic emulsion is made of tiny crystals of silver halide, and after development, clumps of metallic silver (or dye clouds in colour film). Faster, more sensitive films use larger crystals, which is why a 400 or 3200 speed film looks grainier than a 100 speed one. Because the clumps are random, grain shows up as brightness texture rather than colour speckles.
Digital noise has a different cause. Light arrives as individual photons, and the count in each pixel varies randomly: if a pixel expects N photons, the spread is about √N, so dim areas, where N is small, are relatively the noisiest. Add the sensor’s own read noise and the sum is close to a bell-shaped, Gaussian distribution, which is why Gaussian noise is the standard model. The Box–Muller method used here to turn uniform random numbers into Gaussian ones was published by George Box and Mervin Muller in 1958.
About This Tool
This tool adds Gaussian or uniform noise of a chosen standard deviation to a photo, or builds a plain noise texture, in mono or per-channel colour and at any grain size. A seed makes every result repeatable, and the readouts measure the noise actually added, including what was lost to clipping at black and white.
Images are decoded and saved in your browser by the engine behind the Universal Image Converter, so TIFF, PSD, HEIC and RAW files work too. Noise is always generated at full size; the 100% crop shows it as it will be saved. Nothing is uploaded.
Related tools: Floyd-Steinberg Dithering, Glitch Art Generator, and Image Histogram Viewer.
Frequently Asked Questions
What does Amount σ mean?
It is the standard deviation of the noise in brightness levels (0–255). With Gaussian noise about 68% of pixels move by up to ±σ, 95% by up to ±2σ and 99.7% by up to ±3σ. At σ = 12 a mid-grey of 128 mostly lands between 116 and 140. Uniform noise with the same σ is spread evenly over ±σ√3, so ±20.8 at σ = 12.
Mono or colour noise?
Mono adds the same random offset to red, green and blue, so it changes brightness only and keeps the colours, which is how film grain looks. Colour gives each channel its own offset, which produces red, green and blue speckles like a digital camera at high ISO in low light.
Why use a seed?
The noise comes from a seeded random number generator, so the same seed, settings and image always give exactly the same grain. That lets you re-create a result later, match the grain across a set of photos, or change the seed until you like the pattern.
Can noise really hide colour banding?
Yes. A smooth gradient stored with 256 levels can show visible steps, especially after compression. Adding a little noise (σ around 1–3) breaks the straight edges between steps into a fine random texture the eye averages out. This is the idea behind dithering.
Is my image uploaded?
No. The noise is generated and added in your browser, and the image is decoded and saved by the same engine as the Universal Image Converter, so TIFF, PSD, HEIC and RAW files work too. Nothing leaves your device.
How do I use the Image Noise Generator?
Simply pick your options and read the result, which refreshes the instant you change something. There is nothing to submit and nothing to wait for.
Is it free? Does it work without internet?
Yes to both. It is free with no sign-up, and once the page has loaded it keeps working even with no internet.
Where does my data go?
Nowhere — every calculation runs on your own device. Nothing you enter is uploaded, logged, or stored.
Common Use Cases
Film look for digital photos
Mono Gaussian grain at σ 10–14 with grain size 2 gives a 24 MP digital photo the texture of 400-speed film.
Fix banding in gradients
Uniform noise at σ 2 hides the steps in a dark sky or a 1920×1080 gradient wallpaper before it is compressed.
Textures for design
A 512×512 mid-grey noise texture at σ 40 works as an overlay, a paper texture or a bump map.
Testing denoisers
Add colour noise of a known σ with a fixed seed, then measure how much a denoising tool removes.
Last updated: