Generative AI, Art Authentication and Forgery

September 2023

An exploration of the potential role generative AI models could offer art forgers in the twenty first-century.

A Midjourney generative AI rendering of a painting by Pierre-Auguste Renoir, used only from a general online data set.
Giacomo, Researcher & Business Development Manager

Interested in authenticating your painting?

Key Takeaways:

  • Art forgers have been underestimated both in their ability to evade analogue painting analyses but also in their acuity to produce highly persuasive original compositions, a notion that assumes particular significance in the face of generative AI.
  • Generative AI, such as deep learning models like GANs (Generative Adversarial Networks), can produce art that closely mimics the formal, compositional and colouristic characteristics of artists. Widely available generative AI models exist where general data sets of artworks created by old master and modern artists can be used to train a generative AI model, but these could become increasingly powerful with refined data sets of a single artist’s oeuvre.
  • If Midjourney, Open AI's DALL-E or Stability AI's Stable Diffusion can create reasonably convincing works like, for example, the composition in the style of Pierre-Auguste Renoir from a general data set (figure 1), one can only imagine the possibilities opened up by training generative AI models exclusively on data from a single artist that it is replicating
  • Generative AI enables individuals to create new synthetic data which, in the context of art forgery, could serve as a vital tool for forgers to shift beyond the realm of amateur, sloppy pastiches in the production of forged artworks.

Generative AI and the possibilities it offers art forgers crystallise the importance of analogue and digital AI analyses offered by Hephaestus Analytical to ascertain a conclusive authentication result of a given artwork.

Although examples do exist of forgers copying directly from bona fide authentic works which subsequently appear on the market, a useful example being the May 2000 sales at Christie’s and Sotheby’s where two of the same paintings of a vase of lilacs by Gauguin were put to auction at the very same time as discussed in our recent insights article, pastiches are a more common practice for skilled art forgers, a practice that generative AI could radically improve. Generative AI operates on the principle of learning from typically large datasets and then generating new content based on the patterns and features it has learned.

Figure 2: Vladimir Tatlin, Model, 1913. Oil on canvas. State Tretyakov Gallery, Moscow.

In a recent IFAR journal article by Konstantin Akinsha discussing the pervasive issue of forgery in the Ukrainian and Russian Avant-Garde market, the art historian noted the central role that pastiches play for art forgers, referring to Vladimir Tatlin’s bona fide authentic Model (1913) (figure 2) from the State Tretyakov Gallery, Moscow and its role in informing subsequent pastiche forgeries. The linear construction of the model’s nude body, the ochre hues of their skin, the figure’s red fabric support and the abstract treatment of the background informed two forged paintings. Model (1913) (figure 3) is one example, a forged pastiche of Tatlin’s authentic Model which presents the same gleam of white light that crosses the figure’s radically simplified physiognomy to their upper torso from the authentic work, a motif continued in the forged Model (n.d) from the collection of Igor Toporovsky. Known composite frauds often evidence not only formal mimicry but oftentimes amateurish anachronisms. A work previously attributed to Natalia Goncharova, Man on a Scooter (figure 4) (c.1913-1914), included in Denise Bazetoux’s 1994 catalogue raissonné of Goncharova (no. 772), for example, draws upon an authentic painting of a Biker (1912-13) (figure 5) but the central figure uses an adult kick scooter as opposed to a bicycle – a mode of transport designed in 1973, approximately sixty years after the supposed date of the painting’s production.

Figure 3: Previously attributed to Vladimir Tatlin, Model, 1913. Oil on canvas. Private collection.

In highlighting these few examples of pastiche forgeries it becomes increasingly clear not only that forged works have been accepted as part of artists’ oeuvres but also that such known forgeries are characteristically amateur, characterised by historical, stylistic and in many cases linguistic inaccuracies. Generative AI offers forgers the ability to assimilate a data set of museum works and produce new, synthetic compositions from which to paint. Generative AI has the possibility to enable forgers to move beyond mere formal mimicry and, as such, create highly convincing compositions that might be able to fool not only major art market participants but also learned connoisseurs of a given artist to a radically greater extent than ever before.

Figure 4 : (left) Previously attributed to Natalia Goncharova, Man on a Scooter, c. 1913-1914 (Bazetoux no. 772, La trotinette). Private Collection. Figure 5: (right) Natalia Goncharova, Biker (Cyclist), 1912–13 Oil on canvas. Private Collection.

Forgers might use generative AI to not only create entirely new artworks or statistically assess the probable appearance of ‘lost’ works but also given the ease at which a forger might be able to assume an invariably indiscriminate data set of museum works alongside the speed at which generative AI networks operate, it could be possible to produce hundreds of forged compositions, share these digitally and only laboriously produce the forged work once market or academic interest has been expressed. For example, using a 3D visualisation software such as Autodesk 3DS Max alongside a modern rendering engine like Corona 10, an enterprising forger could be able to convincingly display a physically non-existent forged artwork in a digital context.

Although sloppy pastiches pervade the old master and modern art markets, the tools at the disposal of contemporary forgers are incomparably powerful and, as such, highlight the vital role of both analogue and digital art authentication analyses in the arms race against forgery.

Hephaestus Analytical is an art authentication company based in London, UK and New York, USA that offers both private and institutional collections the highest evidentiary standards in the authentication of modern and old master artworks.

Other Posts