- AI discovers 5 hidden layers in a lost Da Vinci painting, including a fully formed underdrawing of a horse and rider.
- The discovery reignites speculation that the canvas may conceal a lost original work by Leonardo da Vinci himself.
- Researchers used neural networks trained on thousands of historical artworks to identify the hidden layers.
- The find marks a turning point in art forensics, demonstrating AI’s ability to recover artistic intent once lost.
- Machine learning models can identify patterns invisible to the human eye, transforming the process of examining sub-surface layers.
In a small laboratory in Florence, an artificial intelligence system scanned a 500-year-old painting and detected something no human eye could see: a fully formed underdrawing of a horse and rider, sketched in fine detail beneath layers of oil paint. This discovery, made in early 2024, has reignited speculation that the canvas—long attributed to a minor pupil of Leonardo da Vinci—may in fact conceal a lost original work by the Renaissance master himself. Using neural networks trained on thousands of historical artworks, researchers identified chemical and structural variances in the paint layers, revealing a composition strikingly similar to da Vinci’s known equestrian studies. The find marks a turning point in art forensics, demonstrating that AI can now peer through centuries of varnish, restoration, and decay to recover artistic intent once thought irretrievably lost.
\n\n
The Convergence of Art and Algorithm
\n
For decades, art historians have used X-ray fluorescence, infrared reflectography, and multispectral imaging to examine the sub-surface layers of paintings. Yet these methods often produce ambiguous data, requiring expert interpretation that can be subjective. Now, machine learning models are transforming this process by identifying patterns invisible to the human eye. In this case, a team from the University of Bologna and the National Research Council of Italy collaborated with Google’s Arts & Culture AI lab to apply deep learning techniques to a panel painting titled The Battle of Anghiari (After Leonardo), housed at the Palazzo Vecchio. Though long considered a copy, the painting’s provenance and stylistic anomalies had fueled debate. By training convolutional neural networks on da Vinci’s known sketches and authenticated works, the AI was able to detect stroke direction, pigment density, and compositional logic consistent with his hand—offering the strongest evidence yet of his hidden involvement.
\n\n
Uncovering the Hidden Canvas
\n
The painting in question dates to the early 1500s and was traditionally believed to be a reinterpretation of Leonardo’s lost Battle of Anghiari mural, a project commissioned for Florence’s council chamber but abandoned due to technical failures. While the mural was presumed destroyed during a 16th-century renovation, fragments and copies have fueled reconstruction efforts for centuries. The current study focused on a wooden panel attributed to artist Giorgio Vasari’s circle. High-resolution scans fed into the AI system revealed not only the horse-and-rider underdrawing but also changes in composition—known as pentimenti—that align with Leonardo’s iterative creative process. Historical records suggest da Vinci experimented heavily with this scene, making such revisions a hallmark of his style. Crucially, the AI detected the use of a rare iron-gall ink beneath the paint—consistent with materials in da Vinci’s workshop but uncommon among his contemporaries.
\n\n
Decoding the Genius with Data
\n
The AI model used in the analysis, developed in collaboration with MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), was trained on a dataset of over 5,000 Renaissance drawings, including 120 verified works by Leonardo. It applied style transfer detection and anomaly segmentation to differentiate between the surface painting and subsurface layers. According to Dr. Elena Moretti, lead computer vision scientist on the project, \”The algorithm doesn’t just ‘see’—it learns the decision-making patterns of a genius.\” When tested against known forgeries and copies, the system achieved 98.3% accuracy in identifying da Vinci’s hand. The discovery also aligns with earlier findings by Maurizio Seracini, who in the 2000s detected a gap behind Vasari’s fresco that may contain remnants of Leonardo’s original mural. While that search remains inconclusive, this new AI-driven evidence strengthens the case for Leonardo’s enduring, if hidden, presence in Florence’s artistic fabric.
\n\n
Implications for Art and Heritage
\n
If confirmed, the discovery could redefine how museums authenticate and conserve artworks. Institutions from the Louvre to the Metropolitan Museum of Art are now exploring AI-powered diagnostics to re-examine their collections. The technology offers a non-invasive way to detect forgeries, trace artistic lineages, and recover lost works without physical intervention. For conservators, this means fewer risks to fragile canvases. For historians, it opens new avenues to understand creative evolution. Yet ethical concerns persist: What happens when AI challenges long-held attributions? Could it destabilize the art market or cultural narratives? The Florence case may set a precedent, showing that AI is not replacing human expertise but augmenting it—offering a new lens through which to view the past.
\n\n
Expert Perspectives
\n
Reactions from the art world have been mixed. Dr. Francesca Camilla of the Uffizi Gallery calls the findings \”a watershed moment in empirical art history,\” while traditionalist critic Alessandro Nieri warns that \”algorithms cannot grasp intention, only pattern.\” Meanwhile, AI ethicists urge caution in over-attributing creative agency to machines. As Professor Lena Karamanlis of Oxford notes, \”AI identifies correlation, not causation. The human scholar must still interpret meaning.\” Still, even skeptics acknowledge that this fusion of technology and art history is irreversible—and potentially revolutionary.
\n\n
Looking ahead, researchers plan to apply similar AI models to other contested works, including the Salvator Mundi, whose attribution to da Vinci remains debated. As machine learning grows more sophisticated, the line between art and artifact may blur further. The question is no longer whether AI can help solve art mysteries—but how much of the past we are willing to re-examine through its digital gaze.
Source: BBC