Solutions · Provenance and EU AI Act

Publish AI product images with provenance and EU AI Act disclosure

Generate with Uwear MCP, then assess disclosure and test delivered files. Uwear’s provenance signals support origin checks; they do not determine legal compliance.

Get a demoSet up Uwear MCP
Model wearing a polka-dot tie-neck blouse, full-body front on the Pure White studio, generated with nano-banana-pro through the Product page set template
Front hero from the Product page set template: polka-dot tie-neck blouse on the Pure White studio, generated with nano-banana-pro through the Uwear MCP.

What the job involves

Proposed owners: the system provider, your legal team and publishing operations. Separate provider marking, contextual deployer disclosure and destination requirements.[4][5][6]

What the job involves
OwnerWhat the source says
AmazonPhotorealistic synthetic people in listing and A+ media require a metadata keyword. The rule is not for every AI edit.[4]
Google Merchant CenterRequires generative-origin metadata for AI product images and retention of embedded tags. This applies to ads and free listings.[5]
Adobe Content AuthenticityIts 24 April 2025 public beta announced batches of up to 50 JPG/PNG files. That historical capacity is not a compliance audit or hourly rate.[6]

How teams do it today

  1. Classify the asset and the placement

    Proposed process: identify provider/deployer roles, creation dates and the actual disclosure context. Keep origin evidence, rights records and visible labels separate.[1][2]

  2. Inspect the delivered derivative

    Test the uploaded and served file, not just the master. Cloudflare has a specific Content Credentials retention setting; generic EXIF retention is not equivalent.[10][14]

Requirements and constraints

Requirements and constraints
RuleRequirement or guidance
Law: Article 50 datesArticle 50 has applied since 2 August 2026. Articles 111(4) and 113 limit the 2 December 2026 transition to provider-side 50(2) marking for systems placed on the market before 2 August.[1]
Law: marking and disclosureArticle 50(2) requires machine-readable marking subject to its scope and exceptions. Article 50(4) covers deepfakes; 50(5) requires clear disclosure at first exposure. Assess the placement with your legal owner.[1]
Platforms: AmazonAdd contains-synthetic-performer to XMP dc:subject for photorealistic AI people. This specific tag excludes real people merely altered by AI and images without people.[4]
Platforms: GoogleRetain generative-origin DigitalSourceType metadata. Amazon’s synthetic-person keyword is a different field and does not alone establish Google acceptance.[5]
Platforms: MetaThe 1 June 2026 rollout update places AI info in About this ad. Confirm region and placement; a menu-only indicator does not automatically settle first-exposure disclosure.[9][1]
Platforms: TikTok ShopRequires proactive disclosure for fully or significantly AI-generated content. The verified scope is TikTok Shop, not a universal TikTok auto-label rule.[13]
Standards: C2PASigned provenance assertions support validation of origin/history. A valid credential does not prove product truth, rights clearance or compliance.[7]
Standards: IPTCDigitalSourceType classifies origin, rather than signing it. The canonical generated-media identifier is http://cv.iptc.org/newscodes/digitalsourcetype/trainedAlgorithmicMedia; see the separately cited compositeWithTrainedAlgorithmicMedia term for generative edits.[8][23]
Delivery: CloudflareWith Content Credentials retention disabled, “any existing Content Credentials will always be discarded.” Enable and test the specific transformation route; downstream edits and recompression can remove signals.[10]

Assistant workflow

How do you run this from your assistant?

Proposed workflow: your assistant runs Uwear MCP; your team owns reference permissions, destination checks and publishing.[16][18][1][17]

  1. Generate and retain the delivered master

    Run Product page set from approved adult garment photos. Uwear applies an invisible TrustMark watermark, C2PA origin metadata where the format supports it, and a server-side log keyed to the final asset SHA-256.[16][18]

  2. Approve the Product page set brief

    Product page set makes front, upper-body and back white-studio photos plus a 360 video per garment. Review the avatar, art direction and cost in credits before generation.[16]

  3. Verify the file and apply contextual disclosure

    Use /verify-ai-image or POST https://api.uwear.ai/public/provenance/verify. Uwear has a visible-label option for outputs classified as deepfakes. Your team tests downstream edits, metadata stripping, recompression and actual first-exposure display.[1][17][18]

Model wearing the same polka-dot tie-neck blouse, full-body back camera edit on the Pure White studio, generated with nano-banana-pro through the Product page set template
Back view from the same Product page set run: a camera edit of the front hero, so the model, light, and studio carry over. Generated with nano-banana-pro through the Uwear MCP.

Run it from your assistant

Run this photoshoot from your AI assistant with a Uwear template.

Connect the Uwear MCP server to ChatGPT, Claude, or Codex and ask for the Product page set template. It is a batch workflow that runs once per garment: front, upper-body, and back product-page shots on a white studio background, plus a 360-degree video, from one garment photo. Any Uwear account can load it by name, and the assistant shows a costed brief in credits before generation starts.

Proposed prompt: adapt the inputs and delivery checks to your own tools. The template returns assets; your team owns final selection and publication.

Copy this prompt into your assistant

Use the Uwear MCP to prepare Product page set for these adult garments with our approved avatar, Pure White and nano-banana-pro. Show the costed proposal and wait for approval. Turn the AI reviewer on and return image and video URLs. Our team will retain the delivered masters, classify the outputs and placement, apply required metadata and visible disclosure, and test delivered files with the public verifier. Generation and QA are not legal approval; do not publish.

Set up the MCP server

What one run does, per garment

  1. Generate the front hero

    One full-body front photo of the garment on the template model, in the Basic white photoshoot art direction on the Pure White studio location.

  2. Edit to the upper-body crop

    A camera edit that reframes the front result to an upper-body front frame instead of generating a new image, so the crop comes from the same session as the hero.

  3. Edit to the back view

    A second camera edit of the front result to a full-body back frame, so the product page gets the back of the garment from the same session.

  4. Render the 360 turn

    A 5-second Kling 3 Pro video from the front photo: the model turns a slow, full 360 degrees in place, garment readable throughout.

Outputs per garment

  • Full-body front photo, 2K, 2:3
  • Upper-body front photo, 2K, 2:3
  • Full-body back photo, 2K, 2:3
  • 5-second 360-degree video

The setup the template pins

Template kind
Batch workflow: runs once per garment row
System key
demo_full_pdp_test_white_studio
Art direction
Basic white photoshoot (system art direction)
Location
Pure White (system studio location)
Model
The Uwear system model by default; the assistant can swap in one of your saved models when it proposes the brief
Image model
nano-banana-pro, 2K, 2:3
Video model
Kling 3 Pro, 5 seconds
QA
Optional. With the AI reviewer on, each result is checked against the source product photo and flagged results are retried
Approval
The assistant shows a costed brief in credits before anything generates

Pure automation

Run it without an assistant

  1. Run asynchronous generation

    Use REST jobs with polling or webhooks for pure automation. Production Automation can start on garment-processed, tag-added or outfit-created events, with immutable published versions, idempotent runs and upfront credit reservation.[16]

  2. Deliver through your own systems

    Your service maps result URLs to the destination and checks the served assets. Studio is the manual path: select the source photos and setup, then download the results.[16]

Tools that target this job

Documented features from the linked sources. These are not independent quality tests.

  • Adobe Firefly

    Adobe documents automatic Content Credentials for entirely Firefly-generated pixels. A fit for that generation path; this does not cover every third-party model or edit.[19]

  • Adobe Content Authenticity

    A provenance companion for existing work, not a fashion generator. A fit for separate attribution; current terms and limits were not verified beyond its 2025 beta announcement.[6]

Frequently asked questions

Yes. Connect Uwear MCP, supply garment photos and approved references, and request a costed Product page set proposal. Approve it before generation; your tools handle destination delivery.[16]

The Commission says a deepfake can resemble a plausible person, not only a named individual. Assess the image and context; commercial creative work has no blanket exemption. Article 50 has applied since 2 August 2026.[1][2][3]

No. TrustMark, C2PA origin metadata, the final-hash log and public verification support origin checks. They do not classify every disclosure duty or replace the required visible label.[7][17][18]

Downstream edits, metadata stripping and recompression can remove signals. Test the source and delivered file, the configured CDN route and the visible disclosure. No round-trip acceptance test is claimed here.[10][18]

Uwear uses credits at $0.10 per credit; rates are on /models. QA is automatic and toggleable: when on, an AI reviewer compares results with source photos and retries flagged results.[16]

Yes. Studio is the manual path: choose source photos, avatar and generation setup, then download results. Your team handles destination formatting, disclosure and publication.[16]

Research

Sources and access dates

  1. [1]
  2. [2]
  3. [3]
  4. [4]
  5. [5]
  6. [6]
  7. [7]
  8. [8]
  9. [9]
  10. [10]
  11. [13]
  12. [14]
  13. [16]
  14. [17]
  15. [18]
  16. [19]
  17. [23]

Next step

Connect your catalog to Uwear.

Start with an assistant-led test, then build the API delivery your catalog needs.

Get a demoSet up Uwear MCP