Caravaggio did not paint many pictures, but he inspired thousands. That imbalance, a small documented oeuvre and an enormous afterlife of imitation, is why attribution in Caravaggio studies can feel like detective work conducted in half-light. Few cases capture that better than The Lute Player, a composition so beguiling, and so frequently copied, that it has generated more than a century of rival claims over three closely related versions: the painting in the State Hermitage Museum in St. Petersburg, the former Beaufort painting from Badminton House (often called the ex-Badminton version), and the version long associated with the Wildenstein collection in Paris.

At first glance, all three seem to promise the same thing: a musician absorbed in performance, a still life that turns paint into atmosphere, and a carafe of water that doubles as a miniature lesson in optics. Look longer, though, and the differences begin to matter. Light behaves differently. Anatomy tightens or slackens. Instruments are described with confidence or with uncertainty. Those small divergences are exactly where attribution lives.

In recent years, Clovis Whitfield has pushed the discussion onto firmer ground by combining close looking with provenance reconstruction and technical evidence. In 2025, Art Recognition added a further layer: convolutional neural network analysis trained on verified Caravaggio images and negative examples drawn from Caravaggisti and close followers. The point was not to replace the connoisseur’s eye, but to offer something the field rarely gets in entrenched disputes: a reproducible, empirical tie breaker when experts cannot agree.

What the evidence now suggests is both bracing and clarifying. The Hermitage and ex-Badminton versions are autograph works, painted in Caravaggio’s early Roman period under Cardinal Francesco Maria del Monte. The Wildenstein version, long defended as the prime original by Keith Christiansen (with Denis Mahon’s influential support at the time), is not by Caravaggio’s hand, but a later copy by another painter.

The Hermitage version: the Undisputed Anchor

The Hermitage Lute Player has long been admired as one of the most luminous survivors from Caravaggio’s early Roman years. Its provenance is unusually continuous, documented in the Giustiniani orbit by the seventeenth century and later acquired for the Russian imperial collection, arriving in the Hermitage by the early nineteenth century. That chain of ownership, coupled with the painting’s technical characteristics, makes it the least vulnerable of the three to speculative doubt.

Recent reporting around the wider Lute Player controversy has treated the Hermitage picture as the benchmark, the version few argue against, in part because it provides the “fixed point” against which the other two can be measured.

The Ex-Badminton version: Once Dismissed Now Re-Entering the Record

The ex-Badminton painting has one of the more dramatic modern biographies. It was sold at Sotheby’s in 2001 for £71,000 as a work from the “circle of Caravaggio,” effectively a high-quality derivative rather than an autograph. The price tells you what the market believed at the time.

The reassessment hinges on a convergence of different kinds of evidence. Whitfield’s reconstruction aligns the picture more plausibly with the world of Del Monte and with the early documentary descriptions of a Lute Player featuring the carafe’s reflected windowlight and the tiny, minutely observed details that Baglione singled out. Those are not just decorative flourishes. In Caravaggio, optical logic is often the signature.

Then, in 2025, Art Recognition’s AI analysis returned an 85.7 percent probability of autograph authorship for the ex-Badminton painting. The result was widely reported, including in The Guardian (September 27, 2025) and Artnet News (September 30, 2025), because it did something the field rarely sees in public: it challenged a long-standing expert hierarchy with a quantified counterargument.

The Wildenstein Version: Prestige without Traction

For decades, the Wildenstein Lute Player enjoyed the confidence that comes from prestigious defenders. Christiansen’s 1990 case, coupled with Mahon’s support, gave the painting a powerful institutional tailwind, amplified by high-profile exhibition history. The argument was essentially connoisseurial: the painting looked right, and it seemed to match Baglione’s famous description. But over time, the case has grown harder to sustain because the painting carries friction in every category that matters. Its earlier provenance is fragmentary. Its iconography deviates in ways that are difficult to reconcile with Caravaggio’s Del Monte milieu. And the painting exhibits precisely the kinds of misunderstandings that skilled copyists can introduce: things that look convincing in reproduction yet fail the stress-test of sustained, technically informed looking.

On the computational side, the Art Recognition analysis placed the Wildenstein version at only 38 percent probability of Caravaggio authorship, far below the ex-Badminton and effectively outside the zone where a credible autograph claim can stand without extraordinary supporting evidence. Reports covering the 2025 findings explicitly contrast the ex-Badminton result with the weakened status of the Wildenstein picture.

Why AI Matters Here, and Why It is Not the Story by Itself

The most useful way to think about AI in this case is not as a verdict machine, but as a procedural tool that forces clarity. Traditional connoisseurship excels at recognizing quality, coherence, and artistic intention, but it can also harden into camps when reputations and past arguments calcify. Provenance and technical analysis can narrow the field, but they do not always deliver a single, decisive “therefore”.

AI analysis, when responsibly trained on complete reference corpora and robust negative examples, can function as a tie breaker in exactly these situations. It does not see “genius.”. It sees patterns: micro-behaviors of paint application, structural regularities, and statistical consistencies that are difficult for the human eye to quantify, especially across dozens of works.

In the Lute Player triad, the important point is the alignment. Whitfield’s archival and technical arguments point one way. The neural-network results point the same way. The combined effect is not merely additive. It is stabilizing. It allows the field to move forward with a more coherent hierarchy: the Hermitage and ex-Badminton paintings as autograph works from Caravaggio’s early Roman period under Del Monte, and the Wildenstein version as an accomplished, historically fascinating copy.

That is the quiet promise of this methodology. Not to end disagreement forever, but to make the next disagreement better, more transparent, and harder to sustain on authority alone.

< Back