Methodology

How Veriso analyzes an artwork

Veriso is an AI tool that produces an evidence-based first opinion. Here is, in plain language, what the system looks at, how it reasons, where its data comes from, and where its abilities stop.

Overview: from your photos to the report

Every analysis follows the same steps. No single step decides on its own: each one produces evidence that the next ones check.

Step 1
Photos
Front, back, signature close-up, labels.
Step 2
Visual reading
Medium, support, technique, subject, visible condition.
Step 3
Signature & marks
Letter-by-letter transcription, alternate readings.
Step 4
Cross-checks
Documented artists, chronology, school, museum collections.
Step 5
Market & report
Comparables, cautious range, confidence, next steps.

Image analysis: what the models see

Several vision models read your photos. Google Gemini performs the detailed visual reading and transcribes inscriptions; Google Cloud Vision extracts text from labels and stamps; models from OpenAI and Anthropic (Claude) then write the reasoning and the report from the collected evidence.

Medium and support

Oil on canvas, gouache on paper, print, photograph… and the visual clues behind that reading.

Technique and style

Brushwork, print screens, plate marks, palette, composition — related to a possible period and school.

Subject and place

What is depicted and, when possible, an identifiable location that helps place the work.

Back and labels

Gallery or exhibition labels, stamps, inventory numbers, handwritten inscriptions.

Visible condition

Craquelure, visible restoration, stains, fading — only what the photo actually shows.

Photo quality

When a detail is blurred, cropped or glared, the report says so instead of guessing.

Signature reading: how we identify and validate

A signature is evidence, not proof: it can be added later, copied or misread. Veriso therefore treats it as a hypothesis to test.

1
Transcribe
Letter-by-letter reading, with alternate readings when a cursive letter is ambiguous (an “S” that could be an “L”, for example).
2
Detect the script
Signatures in Cyrillic, Greek or Arabic script are detected and transcribed in their original script before searching.
3
Search
Readings are compared against documented artist authorities (Getty ULAN, Wikidata, our artist index) and visually against a library of reference signatures.
4
Validate
A name is kept only if it fits: life dates compatible with the apparent period, nationality, and the artist's usual media and subjects.

Attribution levels

Veriso uses the vocabulary of auction catalogues. The level chosen depends on the available evidence — and it directly affects value.

Strongest attributionweakest

Wording

What it means

Typical evidence

Autograph (“by”)

The work is considered to be by the artist's own hand.

Consistent signature, matching style, subject and chronology, provenance or documentation. Veriso never asserts this from photos alone.

Attributed to

Probably by the artist, with some doubt.

Several consistent clues but no independent confirmation.

Studio / Workshop of

Made in the artist's studio, possibly under their direction.

Very close style, different execution, known studio context.

Circle of

By a close contemporary directly influenced by the artist.

Same period and manner, without a direct link to the artist's hand.

School of / Follower of

In the style of the artist or a movement, by a pupil or later painter.

Recognizable style, but period or execution don't match.

After / Manner of

A copy of a known work, or a pastiche in the artist's manner.

Known composition reproduced, or style imitated at a later date.

Unknown / Unattributed

The evidence does not support naming an artist.

No legible signature or unconfirmed reading. The report then places the school and period.

Comparable sales: how the estimate is built

A value range only means something if it rests on real sales of comparable works. Here is how Veriso gathers them.

Auction results

Public sale results from the Millon auction house, ingested and normalized (artist, medium, size, price realized), form a base of comparables for the French market.

Market research

Targeted web research on the artist adds recent public results when they exist.

Marketplace sales

For decorative or entry-level works, completed sales on online marketplaces give an order of magnitude.

Adjustments

The estimate accounts for attribution level, medium, size, subject, condition and provenance. The less certain the attribution, the wider the range.

When data is thin

If there are too few reliable comparables, the report says so and widens the range — or gives none — rather than inventing precision.

Uncertainty handling: what the confidence indicators mean

Every hypothesis in the report carries a confidence level and the evidence behind it. Read it like a weather forecast: it tells you how solid the evidence is, not a guarantee.

High

Several independent clues agree (legible signature, consistent artist, style, chronology, comparables). Still to be confirmed by a specialist for any official use.

Medium

The clues point the same way but one is missing or ambiguous (partial signature, few comparables).

Low

A plausible but fragile hypothesis: uncertain reading, insufficient photos or contradictions. Treat it as a research lead.

What lowers confidence: blurred or glared photos, no back view, partial signature, a little-documented artist, a mismatch between style and dates, scarce comparables. The easiest way to raise it: add a sharp close-up of the signature and a photo of the back.

Limitations: what an AI analysis does not replace

Veriso works from photos. Some questions can only be settled with the work in hand.

Authentication: only a recognized expert, committee or catalogue raisonné holder can authenticate a work.
Physical examination: pigments, canvas, paper, UV light, X-ray and lab analysis are beyond what a photo can show.
Sophisticated fakes: a good copy or forged signature can fool a visual analysis.
Provenance: Veriso reads visible labels and inscriptions, but does not check archives, invoices or registers.
Official uses: insurance, estates, tax or donations generally require a signed appraisal by a qualified professional.

No human specialist reviews reports before they are delivered. Veriso is a first opinion to decide whether a work deserves a full appraisal — and to arrive at the appraiser with a documented file.

Data sources

The sources below feed artist research, museum context and comparables. Availability can vary; when a source does not respond, the analysis continues with the others.

Source

How it is used

Getty ULAN (Union List of Artist Names)

Artist authority: names, variants, dates, nationalities.

Wikidata

Artist identity, dates and short biography.

The Metropolitan Museum of Art — Open Access

Works by the artist held in museum collections.

Rijksmuseum, Art Institute of Chicago, Smithsonian

Public collection data for artist context.

Europeana and Joconde (French museums)

European and French museum collections.