Nous avons dressé une liste des questions les plus fréquemment posées

The first auction house to offer an artwork for sale based solely on a Certificate of Authenticity issued by Art Recognition was Germann Auctions in Zurich, in November 2024. Acceptance of our certificates is determined by each institution independently and may vary depending on its policies, due diligence procedures, and the specific artwork under consideration.

No. Our technology depends on art-historical scholarship and carefully curated reference material. It adds systematic visual analysis at scale to the established pillars of connoisseurship, provenance, and forensic analysis. Every result also undergoes expert quality review.

There must be sufficient reliable visual and scholarly material to build and test an artist-specific model, and the artist’s hand must be meaningfully assessable from an image. The technology is primarily suited to paintings, drawings, prints, and comparable hand-made works. Drip painting, street art, digital or mechanically generated works, sculpture, fall outside the method’s scope. We assess feasibility before accepting a case.

There is no universal standard for the size of a training dataset. Our aim is to build complete datasets by including all validated works of a given artist. As a result, dataset sizes vary widely—some artists have fewer than 50 authenticated works, while others have thousands. The number of negative examples (non-authentic images) typically exceeds the number of authentic ones slightly. These are carefully chosen to closely resemble the authentic works in style and composition.

To address the challenge of limited data, we use both geometric and latent augmentation techniques. Geometric augmentation involves creating variations of the existing images through transformations like flipping, rotating, or mirroring, which helps the AI learn from multiple perspectives. Latent augmentation allows us to generate new training examples in the AI's latent space (a numerical representation of images), effectively expanding the dataset without needing new physical images. Especially thanks to latent augmentation we are now able to train reliable models even on small datasets.

Yes, when the available scholarship and reference material support the distinction. Depending on the artist and the question being examined, the dataset may cover the complete body of work or focus on a particular period, medium, or stylistic phase. Under-documented periods may limit the strength of an analysis.

The probability score represents the AI model’s confidence in its classification. This is known as class probability. The score should be interpreted together with the model-performance information and other context provided in the Report.

The service is not an unrestricted artist-identification search. If you have one or more plausible attributions, we can assess whether the work is stylistically consistent with each proposed artist, provided a suitable artist-specific model can be developed.

Not as part of the AI assessment itself. Provenance, pigment testing, dating, and other physical or chemical examinations are separate pillars of authentication and require the appropriate specialists. The findings from these examinations can inform the selection of artworks and images included in the AI training dataset.

Yes. Please tell us the extent and location of any known or suspected restoration, retouching, overpainting, or other interventions. If available, provide conservation reports, technical images, or an annotated image identifying the affected areas. When restoration or retouching is extensive, we may mask those areas so that the analysis focuses on the remaining parts of the artwork.

No. Art Recognition does not provide appraisal or valuation services. Please consult a qualified art appraiser or relevant market specialist for a valuation.

No. The analysis is conducted from high-resolution photographs, so the artwork normally remains in your possession. There is no transportation, handling, or insurance to arrange.

Please provide clear, sharp images of at least 3000 × 4000 pixels. Photograph the work straight on, with the camera parallel to its surface, using even natural light and no flash. Avoid shadows, reflections, blur, frames, objects, and unnecessary background. A smartphone image can be sufficient if it meets these requirements. Where possible, submit several photographs. Files larger than 15 MB can be shared via WeTransfer or sent to contact@art-recognition.com.

The artwork, ownership information, analysis result, and related communications are treated as strictly confidential.

Yes — when the result is positive. Both products, the AI Authenticity Evaluation and the AI Authenticity Report, deliver a result of the AI authenticity analysis. When the result is positive, we also issue Certificates of Authenticity.

The timeline depends on the artist, the availability and complexity of reliable reference material, the artwork itself, and whether you require an AI Evaluation or an AI Authenticity Report. After reviewing the case, we provide a tailored proposal and timeline before work begins. For collections, multiple proposed attributions, or recurring professional requirements, we can prepare a tailored scope with an estimated timeline.

The fee depends on the artist, the artwork itself, and whether you require an AI Evaluation or an AI Authenticity Report. After reviewing the case, we provide a tailored proposal and fee before work begins. For collections, multiple proposed attributions, or recurring professional requirements, we can prepare a tailored scope and commercial proposal.

Definitely! Write us at office@art-recognition.com afin de recevoir une offre adaptée à votre demande spécifique.