Adobe Trained on 12,000 Photos. The Photographer Got $1,174.

Gerald Carter founded Diversity Photos in 2016 to build a stock archive of diverse representation — real people, real consent, real documentation. In 2023, he discovered that Adobe had ingested his 12,000-image library into both Adobe Sensei and Adobe Firefly without separate authorization or compensation. He spent years attempting to stop it, challenge it, and be compensated for it. He lost on nearly every claim. Adobe’s settlement offer was $1,173.93.

How the Contract Became a Blank Check

Carter signed Adobe’s 2018 Stock Contributor Agreement. The language granted Adobe rights to use images for “developing new features and services.” Adobe argued — successfully — that AI model training constitutes developing new features and services. Generative AI didn’t exist as a commercial technology when Carter signed. The contract didn’t mention training data, machine learning, or artificial intelligence. It didn’t need to.

This is the core legal mechanism that thousands of photographers who contributed to Adobe Stock before 2022 are now exposed to. Broad “future uses” language in pre-AI contracts is being interpreted to cover training data rights that contributors never contemplated and never priced. Carter’s case establishes that Adobe’s position survives at least to summary judgment.

The Arbitration Barrier

Carter didn’t just lose on the merits — he lost on access. Adobe’s contributor agreement routes disputes to arbitration. Arbitration costs are asymmetric: Carter paid $24,000 simply to demonstrate he couldn’t afford to continue arbitrating his negligence claim. The economic design of the dispute resolution process made meaningful challenge prohibitively expensive for an individual contributor, regardless of the underlying legal question.

This is not unique to Adobe. Platform arbitration clauses with cost structures that favor institutional defendants are standard across the stock photography and creative platform ecosystem. Carter’s case makes the cost structure visible in specific dollar terms.

What Adobe’s Response Reveals

Adobe’s $1,173.93 settlement offer for 12,000 professional images — a library built over a decade with documented consent processes — is a pricing signal. It reflects what Adobe calculated as the fair market value of Carter’s contribution to its AI training data, not as photographs sold or licensed, but as training material.

That valuation implies a per-image rate of roughly $0.10 for training data rights. The comparison to commercial licensing rates for the same images — which would typically range from $30 to $300 per image for exclusive use — illustrates the gap between what creative work is worth as licensed content and what AI companies currently argue it’s worth as training data.

The Provenance Infrastructure Response

The legal response to this pattern is slow. The legislative response is slower. The infrastructure response — the camera and software ecosystem building towards content authenticity and opt-out mechanisms — is moving faster than either.

The C2PA standard (Content Authenticity Initiative, backed by Adobe, Canon, Nikon, Sony, and Microsoft) embeds provenance metadata at capture. Future implementations may allow photographers to declare training data opt-out status at the point of image creation — a technical constraint on training data ingestion that doesn’t depend on contract interpretation or arbitration economics. Adobe’s own PhotoShop and Lightroom now support Content Credentials for new images.

The gap is everything created before that infrastructure existed: billions of pre-2023 images in stock libraries, portfolio sites, and platforms that were uploaded under terms that didn’t contemplate AI training. Carter’s 12,000 images are in that gap.

Content created with AI assistance and reviewed for accuracy.