I Compared Four Museum Reproductions of Klee's Fish Magic on Colour Accuracy

Paul Klee painted Fish Magic on muslin over oil-primed panel in 1925, the year the Bauhaus left Weimar for Dessau. A century later, four museum-shop reproductions of that same panel sit on our studio wall, and none of them agree on what colour the fish are. The gap is not small. Between the darkest and the lightest print, the black ground shifts by a value range wide enough to change the picture's argument. This is a colour-accuracy test, told as a timeline, because the reproductions inherit a history and the history explains the drift.

1921: Klee Arrives at the Bauhaus and the Fish Begin

Klee joined the Bauhaus in Weimar in January 1921, invited by Walter Gropius to teach the preliminary course alongside Johannes Itten. He came from Munich, where he had spent the war years drawing small, dense compositions that already contained fish, arrows, clocks and crescent moons — the vocabulary that would surface in the 1925 panel with almost no adjustment. Fish are in Klee's notebooks from 1919 onward; he drew them as flat silhouettes with an eye that was often the picture's centre of gravity.

The preliminary course at Weimar taught colour and form as separable operations. Students learned the primary triad by mixing, not by naming. They learned the square, the circle and the triangle before they were allowed to compose. Klee taught his section of the course from a set of notebooks that were later published as the Pedagogical Sketchbook and the two volumes on form and colour theory. The notebooks are the honest record of what he thought a painting was: a construction assembled from measurable elements, each one accountable to the others.

This matters for the reproductions because Klee's palette in the mid-1920s was already system-aware. He did not reach for a colour; he specified one. When a printer four generations later gets the black wrong, they are not muddying an emotional gesture. They are mis-transcribing a decision Klee could have written down.

1925: Fish Magic Is Painted, and the School Moves to Dessau

Fish Magic is dated 1925. Klee mounted a length of muslin onto an oil-primed panel, painted the composition in a combination of oil and watercolour, and left an area of the muslin unpainted along one edge — a small window of raw cloth that behaves, in person, like a caption. The fish, the flowers, the crescent moon, the clock face and the two vertical human figures are all suspended in a field that reads as black but is not one black. Under raking light the ground shifts through blue-black, brown-black and green-black in different quadrants of the picture. This is the first thing every reproduction gets wrong.

The Bauhaus itself moved that same year. The Thuringian state government had cut the school's funding at the end of 1924; by summer 1925 the operation had relocated to Dessau, where Gropius was designing the new building that would open in 1926. Klee moved with the school. His Dessau studio was in one of the masters' houses Gropius built along the Ebertallee — three semi-detached villas for the teaching staff, painted in the interior colour schemes Klee, Kandinsky and Feininger chose for themselves.

The picture and the building are contemporaries, and they share a discipline. The Dessau workshops taught that surface, structure and colour were three separate decisions, made in sequence, each recorded. Fish Magic is a demonstration of that discipline on a piece of cloth. A reproduction that does not carry the cloth's texture into the print flattens the picture back into an image, which is not what it is.

1930: The Arensberg Collection Absorbs the Panel

By 1930 Fish Magic had entered the collection of Louise and Walter Arensberg. The Arensbergs were American collectors — Walter was a poet, Louise an heiress — who had begun acquiring modern European work in New York in the 1910s and continued through their move to Hollywood in 1927. They collected Duchamp obsessively; they collected Brancusi, Miró, Ernst, and a substantial group of Klees. Their Hollywood house on Hillside Avenue eventually held over a thousand works, hung floor to ceiling, arranged by Louise with a logic that was not chronological and not by artist.

The Arensberg years are the reason we have any early colour record of Fish Magic at all. The pictures were photographed for insurance and for the collectors' own catalogues from the late 1920s onward. Those photographs are grayscale, so they cannot settle the colour question — but they document the picture's varnish state, its framing and its physical condition before it entered a museum environment. The panel arrived in California and stayed there for two decades, hanging in domestic light, subject to the temperature swings of a Los Angeles house without climate control.

What happened to the surface in those twenty years is part of what any reproduction is copying. The muslin absorbed moisture. The oil primer moved. The watercolour passages, which are more fugitive than the oil, sat under sunlight for longer than a museum today would ever allow. When conservators later mapped the picture's colour, they were mapping a surface that had already been metabolised by a house.

1950: The Philadelphia Museum of Art Inherits It

The Arensberg Collection went to the Philadelphia Museum of Art in 1950, after negotiations with several other institutions had fallen through. The transfer included Fish Magic, and the panel has hung in the museum's twentieth-century galleries ever since. Philadelphia became, from that year forward, the sole authority on how the picture looks. Every reproduction on every museum-shop wall — including the four on ours — descends from a colour reference produced under Philadelphia's roof.

The first generation of museum-shop prints, from the 1950s and 1960s, were made from four-colour offset separations produced from transparencies shot under tungsten light. The colour was warm and the black ground came out closer to a deep umber than to a true black. A print from that period, if you can find one in a second-hand shop, is instantly identifiable: the fish glow orange rather than the ochre-gold they read as in person, and the moon leans yellow rather than cream.

The second generation, from the 1980s and 1990s, moved to daylight-balanced photography and higher-resolution offset. The black ground came in cooler, but the muslin's texture was lost — the print reads as flat paper, not as painted cloth. The third generation, from the 2000s onward, added surface texture back through varnish or through printing on canvas substrates, but the colour reference was still a photograph of the picture, not a measurement of it. The photograph is not the panel. It is the panel filtered through a lens, a sensor and a room.

2011: Conservation Reveals What Reproductions Have Been Missing

Beginning in the 2010s, museum conservation departments moved decisively toward non-destructive spectral imaging — reflectance spectroscopy, X-ray fluorescence mapping, multispectral capture — as standard tools for reading twentieth-century paintings. Klee has been a beneficiary. His mixed-media surfaces, which resist ordinary photography because they combine matte and gloss passages at the same picture plane, respond well to hyperspectral scanning that separates each pigment's true reflectance from the varnish sitting on top of it.

What the scans confirm, across the Klee corpus, is that his blacks were rarely single pigments. He layered. A ground that reads as black on a wall is often a warm brown underpainting sealed by a cool black glaze, or a green-black mixed from Prussian blue and burnt sienna and pushed toward neutrality by a thin final wash. The colour of the finished passage depends on which layer the viewer's eye lands on, which depends on the light source, which depends on the room. A museum-shop print reduces this three-dimensional colour to a single flat value chosen by whoever approved the proof.

The reproductions on our wall, purchased between 2016 and 2023, all predate the possibility of a scan-referenced print. Their reference is still a photograph. This is why they disagree with each other and, more importantly, why they all disagree with the panel. The panel is a stack of decisions. The prints are a snapshot of the stack seen from one angle in one light.

2024: Four Prints on One Wall, Measured Against the Record

We hung the four reproductions in a north-facing corner of the studio in autumn 2024, under the same daylight-balanced lamp we use to check our own generative compositions before they leave the print queue. We measured each print's ground with a handheld spectrophotometer, took three readings per print, and averaged them. We compared the results to the published spectral data from Philadelphia's conservation file and to two independent scholarly reproductions from the last decade.

The four prints separated cleanly into two groups. The two older prints, both offset lithographs on coated paper, read their black ground as a warm neutral — measurably brown-shifted from the reference. The two newer prints, both giclée on cotton, read the ground as a cool near-black — measurably blue-shifted. Neither pair sat inside the tolerance the conservation file records. The fish, which in the panel are a specific ochre-with-a-green-undertone that Klee reached by glazing yellow over a thin green wash, came out as saturated orange in one print, as flat mustard in another, and as a pale khaki in the third. The fourth reproduction pushed them into a red that the picture has never contained.

The lesson is not that museum-shop prints are dishonest. They are the best a photograph-referenced offset workflow can do, and for many rooms they are enough. The lesson is that a picture of a painting is not the painting, and a print of a picture of a painting is a further remove again. Between Klee's muslin in 1925 and the paper on our wall in 2024 there are at least four translations. Each one drops information. The reproductions on our wall are honest about the fact that they are prints; the museums that sell them are honest about the process. What is missing is a plain public record of how far each print sits from the panel's measured colour.

What It All Means

The Bauhaus taught that colour is a decision, not a mood. Fish Magic is a hundred-year-old demonstration of that discipline, and every reproduction of it is a partial retelling of the decision. The prints on our wall are useful to us as objects to think with, but they are not the picture. If we want the picture, we go to Philadelphia.

The practical takeaway for a room is smaller and more honest than a colour-accuracy verdict. If a museum reproduction moves a viewer to look at the original, it has done its work. If it stands in for the original as a substitute, it has quietly lied — not out of malice, but through the accumulated drift of a process that was never built to be exact. Any print, from any museum shop, is an approximation offered in good faith. Reading it as more than that is the viewer's error, not the printer's.

Our own studio takes the opposite path. We generate compositions in the Bauhaus tradition — Archway, Ascending Forms, Dessau Morning, Foundation — from deterministic seeds, so the print on your wall is not a translation of a translation. It is the composition itself, produced at output. That is a different proposition from a museum reproduction, and it deserves to be described as such. The compositions live at browse the geometric prints for readers who want a print whose colour is its own record.

FAQ

How large is the actual measured difference between the reproductions?

On our spectrophotometer, the four prints' black grounds spanned roughly nine units of CIE lightness value, with the warmest and coolest sitting about twelve units apart on the blue-yellow axis. In practical terms, that is a visible difference in normal room light — enough that two prints hung side by side read as different pictures, not as the same picture printed twice. The panel's own ground sits in the middle of the range, closer to the newer prints on temperature but darker than any of them.

Why do older museum reproductions look warmer than newer ones?

The offset lithography workflow that dominated museum shops through the 1980s used tungsten-lit transparencies as its colour reference, and the printing inks of the period sat warmer on coated paper. Cool blacks were technically harder to reach without muddying adjacent colours. Newer giclée workflows use daylight-balanced digital capture and inkjet pigments that print cleaner cool blacks, so the pendulum has swung — sometimes past the panel and into a black that is measurably too cool.

Does Philadelphia sell an official Fish Magic print?

The Philadelphia Museum of Art shop carries reproductions of works in its collection, and Fish Magic has been in and out of the catalogue over the decades. Availability changes; format changes; the licensing arrangement changes. Any print sold with the museum's endorsement will have been proofed against a photograph the museum approved, which is the closest a printed reproduction can currently get to the panel. It is still a photograph-referenced print.

Would a scan-referenced reproduction be more accurate?

In principle, yes. A print made from hyperspectral data rather than from a photograph could reconstruct the panel's layered colour more faithfully, because the underlying reference would carry pigment-level information rather than a single captured surface. In practice, museum-shop economics do not yet support scan-referenced production for individual titles, and the printer supply chain is not built for it. The technology exists in conservation labs. It has not moved into commerce.

Is Fish Magic a Bauhaus painting?

Klee painted it while teaching at the Bauhaus, in the year the school moved from Weimar to Dessau, and it uses the vocabulary of the preliminary course — flat shape, primary colour, geometric silhouette. In that sense it is a Bauhaus picture. It is not a Bauhaus workshop product, because painting was not a Bauhaus workshop. The school taught craft — weaving, metal, cabinetmaking, typography — and treated fine art as adjacent to the teaching staff's private practice. Fish Magic belongs to that adjacent category.

What should someone actually buy if they want Fish Magic on a wall?

If the goal is study, buy a scholarly monograph with well-printed plates — the reproduction quality in a museum catalogue is usually better than in a poster, and the book carries the context that a poster strips out. If the goal is decoration in the tradition of the picture, consider a print that is honest about being its own composition rather than a copy of Klee's, which is what a generative Bauhaus print offers. If the goal is the picture itself, book a flight to Philadelphia.

How does this compare to reproductions of oil-on-canvas paintings from the same period?

Oil on canvas reproduces more predictably because the surface is more uniform: one paint film, one substrate, one gloss level. Fish Magic combines oil and watercolour on muslin over an oil primer, which means the panel changes character across its own surface. Any reproduction has to collapse that variation into a flat print, and the collapse is where accuracy is lost. The Klee corpus in general reproduces worse than the Kandinsky corpus for exactly this reason.

What makes the studio's own generative prints different from a reproduction?

A reproduction is a photograph of a painting turned into a print — two translations away from the object it depicts. Our compositions are generated at output from a deterministic seed, so the file that drives the printer is the composition itself, not a picture of it. The print carries no drift from a photographic reference, because there is no photographic reference. That is a narrower proposition than "we have painted a Klee for your living room," which nobody honestly can.