If you left a shutter open for a thousand years, ordinary film would be totally overexposed within the first day and a digital sensor would fill with its own dark current, so the only medium that works is one that keeps responding slowly: paint that fades in light. That is how Jonathon Keats's Millennium Camera works, installed in Tempe, Arizona, in 2015 to finish in 3015. The light-dose maths shows the result: the sun burns a band across the sky in the first days, people and cars vanish completely, and only things that stand for roughly 40 years or more leave a visible ghost.
- Film and sensors saturate within a day; only slowly fading pigment can record for a millennium.
- At a sharp pinhole's f/200, a sunlit scene delivers only 0.13 Mlx·h in 1,000 years, under the 0.22 needed for a just-noticeable fade.
- A full record needs about f/28 for a fugitive colour and f/7 for a madder-like one: bright, but blurry.
- The sun burns its 47° band within days; a building must stand about 40 years to leave a ghost.
- Keats's first Millennium Camera, in Tempe, is due to finish in 3015.
Long exposures are a real genre. Michael Wesely kept pinhole cameras open for up to about 26 months to record the rebuilding of Potsdamer Platz in Berlin from 1997. In 2020 a beer-can pinhole camera at the University of Hertfordshire was found to have been exposing for eight years and one month, recording nearly 3,000 sun trails.
In 2012 Regina Valkenborgh set up a beer can loaded with photographic paper and pierced with a pinhole at the University of Hertfordshire's observatory, to make a solargraph, a picture of the sun's path. She had tried the technique there before, and earlier attempts had often been ruined by moisture and curling paper. This time she simply forgot about it. In 2020 a colleague found the can, still in place. The paper had survived, which she called "a stroke of luck", and when it was scanned it held almost 3,000 trails of the sun across the sky: an exposure of eight years and one month, thought to be the longest ever made. Nobody planned it. The record was set by forgetting.
Kottke: Beer can pinhole camera takes the longest exposure photograph ever (Dec 2020), linking the University of Hertfordshire releaseA thousand years is about 124 times longer than that. Let me work out what light does over that span, and what the picture would look like.
Calculated in this post: 1,000 years divided by the 8 years 1 month (about 8.08 years) of the Hertfordshire beer-can exposure (Kottke, 2020) is about 124.
Why would film or a sensor fail?
Film and sensors fail because they are built to saturate quickly: silver halide film needs only fractions of a second of daylight at normal apertures, and a digital pixel fills its few thousand electrons in milliseconds and then leaks charge on its own (dark current). For a thousand years you need a medium about a billion times slower and stable for ten centuries.
The two laws that decide it
- The Bunsen–Roscoe law of reciprocity says a photochemical effect depends only on the dose, intensity × time, so a dim light for a long time equals a bright light for a short time.
- Reciprocity failure (the Schwarzschild effect): silver film breaks that law at long times, which is why astronomers once hypersensitised plates.
Light fading of colourants obeys reciprocity well: the Canadian Conservation Institute treats fading as proportional to lux × hours. That makes fading pigment a predictable, extremely slow "film".
| Medium | Obeys reciprocity? | How fast it responds | Fit for 1,000 years? |
|---|---|---|---|
| Silver halide film | No, fails at long times | Fractions of a second at normal apertures | No |
| CMOS sensor pixel | Yes, but leaks dark current | Milliseconds | No |
| Fading lake pigment | Yes, closely (CCI) | A day of full sun for the most fugitive grade; centuries behind a small pinhole | Yes |
This museum guideline converts light exposure into fading using the ISO Blue Wool standard, eight reference dyed wools where each grade is roughly 2 to 3 times as stable as the one before. It gives the dose in megalux-hours (Mlx·h) for a just-noticeable fade (ISO Grey Scale 4): with UV present, 0.22 Mlx·h for Blue Wool 1, 3.5 for Blue Wool 4 and 120 for Blue Wool 8; removing UV raises these, to 0.3, 10 and 1,000. "Almost total fade" is taken as about 30 times the just-noticeable dose. Plant and insect dyes are mostly high sensitivity (Blue Wool 1–3); alizarin dyes and lakes are medium (4–6). The limit: these are averages for categories; any single pigment must be tested.
Read the guideline (free)See the data as a table
| Dose to just-noticeable fade ( Mlx·h) | |
|---|---|
| Blue Wool 1 | 0.22 |
| Blue Wool 2 | 0.6 |
| Blue Wool 3 | 1.5 |
| Blue Wool 4 | 3.5 |
| Blue Wool 5 | 8 |
| Blue Wool 6 | 20 |
| Blue Wool 7 | 50 |
| Blue Wool 8 | 120 |
Source: Michalski, Canadian Conservation Institute, Agent of Deterioration: Light (Table 4) · 1 Mlx·h is one million lux-hours. CCI equates 0.17 Mlx·h to about one day of sunlight or one year of museum display at 50 lux.
How does the Millennium Camera work?
Keats's camera is a solid metal pinhole camera. The hole is pierced through a sheet of 24-karat gold, and instead of film the back wall carries oil paint that slowly fades where light falls, so the brightest parts of the scene bleach first and a positive image builds over ten centuries.
The first was installed on 6 March 2015 on the third-level terrace of the ASU Art Museum in Tempe, aimed at the city skyline, with the exhibition of the result scheduled for spring 3015.
Later versions use copper bodies and rose madder, a red lake pigment made from the madder root and used by painters for centuries. More have been placed at:
- Amherst College;
- around Lake Tahoe;
- from 2023, the desert near Tucson.
Keats has said he kept the design simple because anything complicated is liable to break.

The authors made lake pigments by dyeing wool with cochineal, kermes, lac or madder, extracting the dye and precipitating it onto alum, plus brazilwood, alizarin crimson and several yellows. They painted them out in gum arabic, egg tempera or linseed oil, sometimes mixed with lead white, aged them under fluorescent lamps and measured the colour change. Anthraquinone reds, alizarin and madder, were the most stable on any given substrate; tin-based substrates were the least stable; thick, heavily pigmented layers faded less than thin ones. A 2025 follow-up by the same authors found madder the most stable red as both textile dye and lake. The limit: accelerated tests under lamps, not real daylight over centuries.
Read the bulletin (free)How much light would reach the paint in a thousand years?
For a typical pinhole at f/200, the paint behind a sunlit scene receives only about 0.13 Mlx·h in a thousand years, less than a just-noticeable fade even for the most fugitive Blue Wool class. To fully record ordinary buildings, the hole must be much bigger than a normal pinhole. Here is the working, with my assumptions stated.
- Assumptions (my estimates). Average luminance of the daylit scene \( \bar L = 1{,}500 \) cd/m², averaged over seasons, cloud and the low sun of mornings and evenings (a sunlit 20%-reflectance wall at 100,000 lux is about 6,400 cd/m²). Daylight for \( 12 \) h a day, so \( t = 1{,}000 \times 365 \times 12 = 4.38\times10^6 \) h. A pinhole of f-number \( N \).
- Image-plane illuminance from the camera equation (ignoring edge fall-off):
\[ E = \frac{\pi \bar L}{4N^2} = \frac{\pi \times 1500}{4 \times 200^2} \approx 0.029\ \text{lux} \]
- Total dose over a thousand years:
\[ H = E\,t = 0.029 \times 4.38\times10^6 \approx 1.3\times10^5\ \text{lx·h} = 0.13\ \text{Mlx·h} \]That is 0.6 times the 0.22 Mlx·h needed for a just-noticeable fade of Blue Wool 1.
Calculated in this post from the scene luminance, daylight hours and f/200 aperture stated as assumptions in step 1, and the Canadian Conservation Institute's Table 4 (0.22 Mlx·h for Blue Wool 1, UV present).
- Solve for the aperture that finishes the job. For "almost total fade" of a Blue Wool 1 colour, \( H = 30 \times 0.22 = 6.6 \) Mlx·h:
\[ N = \sqrt{\frac{\pi \bar L t}{4H}} = \sqrt{\frac{\pi \times 1500 \times 4.38\times10^6}{4 \times 6.6\times10^6}} \approx 28 \]For a medium-sensitivity colour like a madder lake (Blue Wool 4, 30 × 3.5 = 105 Mlx·h), \( N \approx 7 \).
- The sun is different. The noon sun's luminance is about \( 1.6\times10^9 \) cd/m², a million times the scene. At f/200 its image gets \( \pi \times 1.6\times10^9/(4 \times 200^2) \approx 31{,}000 \) lux, enough for a just-noticeable Blue Wool 1 fade in about 7 hours. The sun's path is burned in within the first days and saturates long before the buildings register.
Calculated in this post from the noon solar disk luminance of 1.6 × 10⁹ cd/m² (Wikipedia: Orders of magnitude (luminance)), the f/200 camera equation above and the CCI dose for Blue Wool 1 (0.22 Mlx·h).
Two sensitivity checks matter:
- Brighter scene. Double the average luminance (a bright desert like Tempe) and the dose doubles, so the right f-number rises by \( \sqrt2 \).
- No UV. Filter out UV and the CCI table says fugitive colours need roughly 1.4 to 3 times more dose.
Either way, the conclusion holds: a sharp pinhole is too slow even for a millennium, so a thousand-year camera trades sharpness for speed or uses a very fugitive colour.
A 10 cm-deep camera at f/28 has a 3.6 mm hole, against an optimum of about 0.23 mm, so its picture is soft, a blur of about 2° per point. I don't know the actual aperture Keats chose; this is my own estimate of the tradeoff any such camera faces.
Use this to check what you just read: change the f-number, the scene brightness or the pigment, and see whether the paint finishes fading in time.
Design a thousand-year camera

What would the thousand-year picture show?
It would show a bleached band where the sun crossed the sky, a brighter sky, and a city in which each building appears with a strength proportional to the share of the thousand years it stood. People, traffic, festivals and even a whole human lifetime would be invisible.
- Mixing rule. With reciprocity, the final fade at a point is the time-weighted average of everything that stood in front of it. An object present for a fraction \( \varphi \) of the exposure, with luminance contrast \( C \) against what replaced it, changes the dose there by about \( \varphi C \).
- Visibility threshold. The eye notices a brightness difference of roughly 1 to 2% in good light; classic estimates of this Weber fraction run from 1/100 (Weber, 1834) to 1/64 (Bouguer, 1760), and real thresholds depend on size, blur and viewing conditions. So an object shows if
\[ \varphi\,C \gtrsim 0.02 \quad\Rightarrow\quad \varphi_{\min} = \frac{0.02}{C} \]
- Plug in. A pale building against a darker hill, \( C = 0.5 \): \( \varphi_{\min} = 0.04 \), or 40 years of a thousand. A person standing still for an hour: \( \varphi = 1/(4.38\times10^6) \approx 2\times10^{-7} \), five orders of magnitude too faint.
Calculated in this post from the 2% visibility threshold above, an assumed value near the historical Weber fractions of 1/100 and 1/64 (Telescope-optics.net), an assumed contrast C = 0.5, and the 1,000-year, 12-hours-a-day exposure assumed in step 1 of the dose working.
- The caveat. Fading isn't perfectly linear: once a colour is mostly gone, extra light changes little. Early arrivals in a fully-faded zone are recorded less than late ones. Treat the 40-year rule as a first estimate.
The sun's band
The sun track is the most predictable part. Each year the sun's path moves between declinations of +23.44° and −23.44°, so a thousand-year image holds a solid band about 47° tall in the sky, the same band each year, laid down 1,000 times.
Earth's axial tilt is shrinking by about 46.8 arcseconds per century, so over the millennium the band's edges creep inward by about 0.13°, a quarter of the sun's own width. The sky is effectively a fixed stamp; the city is the variable.
| Exposure | Length | Medium | What it recorded |
|---|---|---|---|
| Typical phone night mode | a few seconds | CMOS sensor, many frames | Low-light scene |
| Solargraph on photo paper (common practice) | 3 to 6 months | Unprocessed photographic paper, scanned | Daily sun arcs |
| Michael Wesely, Potsdamer Platz, Berlin | up to about 26 months (from 1997) | Pinhole camera, film | Translucent buildings rising |
| Regina Valkenborgh's beer-can camera, Hertfordshire | 8 years 1 month (2012–2020) | Photo paper in a can | Nearly 3,000 sun trails |
| Jonathon Keats, Millennium Camera, Tempe | 1,000 years (2015–3015) | Fading oil paint, gold pinhole | The Tempe skyline, if it survives |

Could the camera itself survive a thousand years?
The optics and paint can; the setting is the risk. A solid copper or steel box with a gold pinhole has no moving parts, gold does not corrode, and oil-bound pigments in thick layers are among the most durable materials in museums.
What threatens the image is people:
- theft;
- demolition of the building it sits on;
- simple forgetting.
A thousand-year camera needs an institution that lasts a thousand years.
The record is instructive: the longest exposure we have was finished by accident, by someone who forgot a can for eight years. Wesely's Berlin pictures needed a building site that would stay unchanged long enough. Keats's answer is to spread copies across several sites and make them simple, so the odds that one survives stay reasonable.
The Mughal miniatures and Ajanta murals in this blog are the evidence that pigments can last centuries when people look after them; see the history of the Ajanta cave paintings and Mughal miniature painting. The shorter, practical version of slow photography is in the shutter speed chart.
A thousand-year exposure is possible, but not on film or a sensor. It works with a medium that obeys reciprocity and fades slowly, such as lake pigments rated by the Blue Wool scale, behind a pinhole far larger than a sharp one: at f/200 the scene delivers only 0.13 Mlx·h in a millennium, under the 0.22 Mlx·h for a just-noticeable fade of the most fugitive class, so a full record needs about f/28 for a fugitive colour and f/7 for a madder-like one.
The sun burns its 47° band in days; buildings appear only if they stood for roughly 40 years or more. The deciding risk is not physics but whether anyone keeps the camera in place until 3015.
What's new
- The camera with the 1,000-year-long exposure, revisited Hyperallergic, Jan 2024 (updated Nov 2025)
- A Millennium Camera starts a 1,000-year exposure of the Tucson desert Smithsonian Magazine, 2024
- Kirby and Saunders find dyes fade less on wool than as lake pigments, with madder the most stable red Heritage (MDPI), Jul 2025
What if next?
- What if the world had only one colour? A rose madder millennium is a one-colour image; here is what one colour would do to everything else.
- What if you painted a rangoli a kilometre long on an Indian highway? Pigment, sunlight and traffic, on a timescale of days instead of centuries.
- What if you took a selfie from a billion kilometres away? The opposite light problem: almost no photons, and the clock running.
The physics of brightness and pinholes is worked in more detail in what if your camera were the size of a building. More from the studio at Kala.
Questions people ask
What is the longest exposure photograph ever taken?
The longest completed one known is a beer-can pinhole image by Regina Valkenborgh at the University of Hertfordshire, exposed for eight years and one month (2012 to 2020). Jonathon Keats's Millennium Cameras are designed for 1,000 years; the first, in Tempe, ends in 3015.
Why do people disappear in long-exposure photos?
Because each point of the image records the time-average of what was in front of it. A person who stands for 1 second in a 100-second exposure contributes 1% of the light at that spot, which is near the edge of visibility; anyone moving contributes far less.
How does solargraphy work without developing the paper?
Photographic paper darkens directly under strong light over weeks (a print-out effect), so a pinhole camera loaded with paper and left for months records the sun's daily arcs. The paper is then scanned without chemical development, because developing would turn the whole sheet black.
Does light damage follow reciprocity?
For fading of colourants, approximately yes: conservators treat damage as proportional to the dose in lux-hours, so 50 lux for a year equals a day of sunlight. Silver halide film does not, and loses sensitivity at very long exposures, which is called reciprocity failure.
Sources
- Michalski, Canadian Conservation Institute: Agent of Deterioration: Light, Ultraviolet and Infrared (2018) (source for the Blue Wool doses, the 30× full-fade rule and dose proportionality)
- Saunders and Kirby: Light-induced Colour Changes in Red and Yellow Lake Pigments, National Gallery Technical Bulletin 15 (1994) (source for madder and alizarin stability)
- Kirby and Saunders: A Comparison of the Fading of Dyestuffs as Textile Colourants and Lake Pigments, Heritage (2025) (source for madder as the most stable red)
- ASU News: ASU Art Museum to document Tempe with history's slowest photograph (2015) (source for the install date, gold pinhole, paint and 3015)
- Hyperallergic: The camera with the 1,000-year-long exposure (2024) (source for later Millennium Cameras)
- Popular Science: Pinhole cameras capturing 1,000-year photos of Lake Tahoe (source for the copper body and rose madder)
- MoMA: Michael Wesely: Open Shutter (2004) (source for Wesely's multi-year exposures from 1997)
- Kottke: Beer can pinhole camera takes the longest exposure ever (2020) (source for 8 years 1 month and nearly 3,000 trails; links the University of Hertfordshire release)
- Wikipedia: Axial tilt (source for 23.44° and the 46.8 arcsecond per century decrease)
- Wikipedia: Orders of magnitude (luminance) (source for the sun's 1.6 × 10⁹ cd/m²)
- Telescope-optics.net: Eye intensity response (source for the historical Weber fractions of 1/100 and 1/64)
- Wikipedia: Reciprocity (photography) (source for the Bunsen–Roscoe law and the Schwarzschild effect)
