Short answer
Product images often show brand names, logos and packaging, and brand questions are not the same as copyright questions. Buyers should ask who can authorize the images, whether the license reaches the brand elements shown, what the model will be used for, and how brand content is handled in delivery. This page gives diligence questions and clause topics. It is not legal advice.
Why this is a separate question
A shelf photo, a packshot or a warehouse tote image is full of marks: manufacturer names, product lines, logos, trade dress such as distinctive packaging shapes and colors, and sometimes retailer private labels. Buyers who review image licenses usually think first about who took the photo and who owns the copyright in it. Brand elements add a second layer. The question is not only who owns the picture but also who owns what is pictured, and what the license lets the buyer do with a model trained on it.
This page does not state what trademark law requires. That depends on jurisdiction, use and facts, and belongs with counsel. Its purpose is to give a procurement or legal lead a list of questions to settle before a purchase, so that brand content does not surface as a surprise at delivery or at deployment. For the copyright side of product photos, see product image rights; for the broader license clauses, see licensing terms for robotics training data. This is not legal advice.
What the available background does and does not cover
The U.S. Copyright Office's report on generative AI training identifies points in development where copying may implicate rights, analyzes fair use, and examines licensing options (U.S. Copyright Office, 2025). It is framed around copyrighted works. It is useful context for the copyright layer of an image license, and this page uses it for nothing else. It is not a source on trademarks, and a buyer should not read it as an answer to any brand question below.
Where brands appear in robot-perception data
| Data type | How brands appear | Why it matters to a buyer |
|---|---|---|
| Studio packshots | Brand and packaging are the subject | Often supplied by manufacturers or retailers with their own terms |
| In-situ shelf or tote images | Many brands in view, partly occluded | Supplier may not control the brands shown |
| Catalog images with text | Product names and marks in captions | Metadata may carry marks as text |
| Video of workers handling goods | Packaging visible in motion | Brand visible incidentally, not the point of the recording |
| Synthetic or rendered assets | Marks recreated by artists or scanned | Origin of the artwork needs checking |
Retail recognition tasks, covered in grocery and retail product recognition, and text on packages, covered in labels, barcodes and text on packages, both put marks in focus. The difference between a brand being the subject and being incidental often shapes the questions. See packshot vs in-situ images for the capture difference.
Questions that separate brand from copyright
- Authority. Does the supplier hold authority over the brand elements, or only over the photograph? A retailer may hold photos of products made by others.
- Source of images. Did images come from manufacturers, a retailer's catalog, a distributor feed, or the supplier's own camera? See retailer vs manufacturer object data.
- Permitted use. Does the license cover training a recognition model that identifies brands, as opposed to one that only detects generic object classes? Brand recognition is a distinct use and should be named in the use grant.
- Outputs. Could the buyer's system generate, display or reproduce marks, for example in synthetic images or in user-facing outputs? If so, say so in the license discussion.
- Third-party agreements. Do retailer or marketplace terms restrict reuse of the images or the marks in them?
- Disclosure. Will the supplier tell the buyer which brands appear, so that conflicts can be checked?
Blurring or masking logos reduces what is visible. It does not settle whether the supplier may license the images, and it can change the data's value, since a model that has never seen a logo cannot learn to read one. Decide deliberately whether brand masking is a requirement or a harm. This applies equally to any removal of names or identifiers: it changes content, not permissions.
Clause topics to discuss
The following are topics for a license draft. They are not recommended wording and no clause is implied to be sufficient.
| Topic | What to settle |
|---|---|
| Scope of images | Which collections, dates and sources are covered |
| Rights warranty | What the supplier states about its authority over images and visible marks |
| Brand-bearing content | Whether the license extends to brand elements shown, or the supplier excludes them |
| Permitted uses | Training, evaluation, internal use; whether brand recognition is included |
| Outputs | Whether model outputs may include marks, and who bears responsibility |
| Notice and takedown | How a brand owner's objection is handled, and what happens to trained models |
| Deletion and audit | What the supplier may request and what the buyer can document |
| Indemnity and liability | Allocation if a claim arises; for counsel to negotiate |
A buyer's counsel decides which of these matter for their use. For plain-language background on license structure, see the data license terms explained guide and SourceX's licensing terms page.
Illustrative image manifest fields
This Illustrative manifest shows fields that help a buyer track brand content, using synthetic values.
# Illustrative only - not an available dataset
image_manifest_v0:
image_id: img_093311
source_type: retailer_catalog # manufacturer_feed / own_camera / third_party
capture_context: packshot # in_situ / video_frame / render
brands_visible: [brand_017, brand_022] # pseudonymous placeholders
brand_is_subject: true
supplier_authority_statement: covers_photograph_only # example value
masking_applied: none # or logos_masked
usage_grant_ref: grant_A # links to a license clause set
restrictions: [no_redistribution, no_output_reproduction]
Recording supplier_authority_statement per image or per collection makes gaps visible early. A manifest does not create rights; it documents what the supplier claims.
What this page cannot settle
It cannot tell you whether a particular use infringes anything, whether a license is sufficient, or how a court in any jurisdiction would treat brand elements in training data. It does not interpret trademark law, and the single source cited concerns copyright. It also cannot make an image set suitable for manipulation: product images support recognition, detection and pose work, and do not show how objects behave when handled, so humanoid control needs robot or teleoperation data. See object recognition data vs manipulation training data.
Buyers should also note that many robot-perception tasks need only generic classes or instance identity without brand names. Where that is enough, specifying it in the use grant keeps the license narrower.
Questions to ask suppliers
- Who took or produced the images, and under what terms were they obtained?
- Do you hold authority over the brand elements visible, or only over the photographs?
- Which brands appear most often, and can you list them?
- Are any images supplied by manufacturers, retailers or marketplaces under terms that restrict reuse?
- Have any brand owners objected to use of their images or marks?
- Is masking of logos available, and what does it do to the rest of the image?
- Can you confirm in writing which uses of brand-bearing images your license permits, including brand recognition?
- How would a takedown request be handled after delivery?
Explore sourcing and licensing
If you are licensing product images and want brand questions raised early, describe the image types, brands and permitted uses to SourceX. SourceX looks for US businesses that may hold suitable material, checks the data and the supplier's licensing permissions, agrees pricing and allowed uses in a license, and coordinates delivery. A request does not guarantee a match; SourceX reviews the requirement and follows up about potential sources. Start at the humanoid training data guide or the product and object data hub. The catalog page on images and inspection photos describes the general asset type.
Explore sourcing and licensing options with SourceX
Sources
- Copyright and Artificial Intelligence, Part 3: Generative AI Training (pre-publication version) — copyright background for AI training; not a source on trademarks.