Is It an Artemisia? Looking Beneath the Surface at the Galleria Spada

Artemisia Gentileschi is now among the most celebrated painters of the seventeenth century. Her dramatic subjects, forceful heroines and intensely physical treatment of paint have made her an art-historical icon, while the events of her life have attracted almost as much attention as her pictures. Yet the renewed enthusiasm for Artemisia has also made the attribution of paintings to her increasingly consequential. Every accepted work can reshape our understanding of her development, her workshop and her relationship with her father and teacher, Orazio Gentileschi.

In this episode of Is It?, we considered two paintings that have belonged to Rome’s Galleria Spada since the seventeenth century: Saint Cecilia et Madonna Nursing the Christ Child. Both entered the collection in 1636 and were recorded early on as works by Artemisia. They were not originally commissioned or purchased by the Spada family. Instead, they arrived as part of the property of Maria Veralli, who had received them from a tenant, Alessandro Biffi, in payment of unpaid rent. It is a pleasingly mundane route into one of Rome’s great Baroque collections: two paintings now regarded as museum masterpieces once functioned as settlement for a landlord’s debt.

The authorship of Saint Cecilia has attracted relatively little controversy. The Madonna and Child has had a more complicated history. Although a seventeenth-century inventory identified it as Artemisia’s work, the formidable connoisseur Federico Zeri later attributed it to Giovan Francesco Guerrieri. Zeri was rarely careless, and his opinion could not simply be dismissed. But even the sharpest eye can be misled when only the visible surface is available for inspection.

To reconsider the two paintings, I was joined by three guests representing complementary forms of expertise. Adriana Capriotti, director of the Galleria Spada, brought the paintings’ history and their place within the collection. Technical art historian Beatrice de Ruggieri discussed the X-rays, infrared imaging and material investigations conducted by her laboratory, M.B. Diagnostica Artistica. Alessandra Conti of Art Recognition explained how the company’s AI model was trained to identify Artemisia’s hand and revealed its analysis of the two works.

The result was not a simple verdict delivered by one method. It was something more interesting: a demonstration of how different kinds of evidence can clarify one another.

Paintings have lives beneath their visible surfaces

When we stand before a painting, we instinctively treat what we see as the artwork. Yet the final surface is only the last stage of a process. Beneath it may be abandoned figures, altered poses, reused compositions and evidence of decisions that the artist eventually concealed.

The technical examination of the Spada Madonna and Child revealed that Artemisia had originally painted a substantially different version of the subject. X-rays showed changes in the positions of both the Virgin and the Christ Child. The underlying composition closely resembles another Madonna and Child by Artemisia in the Palazzo Pitti in Florence, although the Pitti painting has generally been dated later on stylistic grounds.

This produces an apparent chronological puzzle. How could the supposedly later composition already exist beneath an earlier painting?

The answer may lie in workshop practice rather than in a simple linear development from one finished composition to another. Beatrice’s analysis found lines suggesting the use of a cartoon or tracing method. Orazio Gentileschi is known to have reused cartoons, figures and compositional elements. Artemisia learned in his workshop and may have inherited not only his techniques but also a stock of drawings and patterns that could be adapted at different points in her career.

The hidden image also appears closely related to Orazio’s Madonna and Child in Budapest. Adriana suggested that the underlying composition may reflect the work of a very young Artemisia, around 1610–12, still painting under the powerful influence of her father. The two artists may even have worked from the same cartoon.

This matters because it complicates the idea of an artist’s “hand.” We often imagine that an artist possesses a stable and immediately recognizable style. In reality, style is learned, borrowed, revised and sometimes deliberately abandoned. A young painter may work so closely within a teacher’s visual language that separating the two becomes difficult even when the authorship is documented.

The technical images allow us to witness Artemisia separating herself from Orazio. She began with a composition rooted in his workshop practice, then substantially rethought it. The visible painting is therefore not just an image of the Virgin and Child. It is a record of artistic development.

The two paintings reveal different ways of working

The examination of Saint Cecilia produced a very different picture of Artemisia’s process.

Unlike the Madonna and Child, the composition does not contain major changes. The principal figure appears to have been planned with confidence, probably with the help of a detailed preparatory drawing. Technical imaging showed that the lute was painted over an already completed portion of the figure. The instrument was therefore added after the body and costume had been established.

Beatrice proposed that Artemisia may have painted the figure directly from a model, or made careful studies from a model before transferring the design. There are small alterations in the folds of the drapery, but no wholesale reconstruction of the composition.

Placed beside one another, the two works give us contrasting views of the same artist. In Saint Cecilia, Artemisia appears assured and methodical. In the Madonna and Child, she revises, repaints and negotiates with an inherited composition.

This is precisely why technical analysis is so valuable. Had we examined only the finished surfaces, we might have concluded that the paintings’ differences indicated different authors. Once their internal structures are visible, another interpretation becomes possible. They may instead represent different circumstances, stages of development or methods of construction.

Attribution studies sometimes suffer from the expectation that every picture by an artist should resemble every other picture. But artists do not paint under laboratory conditions. Their work changes according to date, location, subject, available models, workshop assistance and the demands of patrons. Artemisia’s early dependence on Orazio is especially significant because the difficulty of separating their hands is not merely a problem for present-day scholars. It is part of the historical reality of how she learned to paint.

The evidence beneath the surface does not weaken the attribution of the Madonna and Child. It explains why the visible surface is unusual.

The AI results became meaningful only when interpreted

Art Recognition’s model returned strikingly different results for the two paintings.

For Saint Cecilia, the model calculated a 96.9 percent probability of authenticity. That is an exceptionally strong result and corresponded with the painting’s secure historical attribution and the technical evidence of a coherent working process.

Le débat autour de Madonna and Child received a probability of 57.1 percent. Considered in isolation, this appears much less conclusive. In previous Art Recognition studies, securely attributed works have often scored around 65 percent or above, although percentages must always be interpreted within the circumstances of the individual model and object.

The most revealing part of the analysis was not simply the overall number but the areas that influenced it. One significant region was around the Virgin’s elbow and sleeve—precisely where the technical investigation had revealed substantial alteration. Alessandra also noted that the design of the sleeve is unlike Artemisia’s usual treatment of costume and may therefore have confused the model.

Training an AI to recognize Artemisia presents further difficulties. Compared with a prolific modern artist, the number of uncontested works available for training is limited. Art Recognition constructed a contrast set containing works by thirteen related artists, as well as imitations and synthetic images. Yet Artemisia’s closeness to Orazio caused a particular problem: in early testing, the model confused her work with her father’s.

That confusion is not necessarily a failure. It may be historically accurate.

If the Spada Madonna was painted when Artemisia was still young, began from a composition closely related to Orazio’s and then underwent extensive revision, we should not expect it to resemble a straightforward example of her mature style. The result of 57.1 percent may register exactly the complication revealed by the X-rays: Orazio lies, figuratively and perhaps compositionally, beneath Artemisia’s painting.

AI image analysis did not overturn the historical attribution. Nor did the documentary record make the AI result irrelevant. Instead, the apparent disagreement encouraged a closer investigation. Once the technical findings, the early date and the father-daughter workshop relationship were taken into account, the lower score became intelligible.

This is an important distinction. Authentication technologies are sometimes presented as machines that will replace the judgement of scholars. The Spada paintings demonstrate the opposite. The model had to be trained using decisions made by art historians. Its result then had to be interpreted by experts who understood Artemisia’s career, Orazio’s workshop and the physical construction of the paintings.

At the same time, the human eye alone had once led Federico Zeri away from the documented attribution. The lesson is not that connoisseurship is obsolete, but that no single method should be expected to carry the entire burden.

The most compelling authentication work is collaborative. Documents tell us how the paintings entered the collection and what they were called in the seventeenth century. Technical imaging reveals hidden compositions and working methods. AI compares visual features across a body of work with a consistency unavailable to the unaided eye. Art historians bring these strands together and determine what they mean.

There is an appealing irony here. A technology often described as futuristic helped illuminate a very traditional artistic relationship: a daughter learning from her father, using the same workshop materials and gradually forming a hand of her own.

Saint Cecilia looks like an Artemisia and produced an emphatic AI result. The Madonna and Child is more complicated, but its complications are what make it so revealing. Rather than presenting a weaker version of Artemisia, it may preserve the moment at which she began to become herself.

You can listen to the whole episode on Spotify or Apple podcasts.

Beneath the surface: a visual guide

Technical examination by Emmebi Diagnostica Artistica, supported by Artemisia Gold. Images and diagnostic descriptions sourced from Artemisia Gold’s project page.