Resolution, Licensing & Usage Rights in AI Product Imagery
By the ORA Lab team · Updated 28 August 2026 · 9 min read
Key takeaways
- Three separate questions get tangled under 'can I use this image?': contract rights (what your tool's ToS grants), copyright status (what the law protects), and platform rules (what channels accept). Clear them one at a time.
- Most commercial tools grant full commercial usage of outputs at paid tiers, but 'you can use it' and 'you own enforceable copyright in it' are different claims; purely AI-generated images have weak copyright protection in most jurisdictions.
- The clauses that actually bite: rights lapsing on cancellation, AI-model likeness terms for on-model imagery, training-use of your uploads, and free-tier restrictions teams forget they started under.
- Resolution needs are channel-math, not vibes: marketplace zoom wants 1,600 to 3,000px on the long edge, social crops far less, print far more, buy the tier that ships what your biggest channel needs.
- Your product photos entering the tool matter as much as images leaving it: confirm your uploads stay yours and check whether they feed model training by default.
Somewhere between generating a beautiful campaign image and running it as a Diwali ad sits a stack of unglamorous questions nobody enjoys: does the tool's licence actually cover paid advertising? Who owns this image, and can anyone own it? Will it survive marketplace zoom at 1,600 pixels? If we cancel the subscription in March, do the images come down? Teams usually discover these questions in the wrong order, after the campaign is live, because the answers hide across a ToS, a copyright doctrine, and a platform spec that no one document connects.
This guide connects them. One disclaimer worth stating plainly: we build AI imagery systems, we read a lot of terms-of-service, but this is orientation, not legal advice, for contracts of real consequence, have counsel read the actual ToS you're signing. What we can do is show you exactly where to look and what the clauses mean in commerce terms.
Three questions wearing one trench coat
'Can I use this image?' is really three questions. Contract: what usage rights does your tool's terms-of-service grant you? This is the one you control by choosing tools and tiers. Copyright: what does the law protect and who holds it? This one is set by jurisdiction and largely out of your hands. Platform: what do Amazon, Meta, Google, and the marketplaces accept? This one is channel policy, and it moves independently of the first two. An image can clear all three, or clear the contract while failing the platform. Teams that keep the three separate make clean decisions; teams that blur them either over-restrict (leaving usable imagery on the table) or under-check (finding out during a dispute).
Contract rights: what the ToS actually grants
- Commercial-use grant: paid tiers of mainstream tools generally licence outputs for full commercial use, listings, ads, packaging. Free tiers often don't, and teams frequently generate their first real assets on the trial tier they forgot they're on.
- Survival after cancellation: the clause that matters most and gets read least. Rights to already-generated images should survive subscription end ('perpetual licence to outputs'); if usage is conditioned on an active subscription, your back catalog has a landlord.
- Your uploads: confirm your product photos remain yours and check the training-use default, reputable tools offer an opt-out for using your content to improve models; luxury brands with unreleased collections should treat this as a hard requirement.
- Exclusivity of outputs: most tools don't promise your generated image is unique to you (similar prompts can produce similar images for others). Usually harmless for product-anchored imagery, your SKU is the differentiator, but relevant for brand-defining campaign art.
- Indemnification: enterprise tiers increasingly offer IP indemnity for outputs; self-serve tiers almost never do. If your legal team asks 'who stands behind this image?', this clause is the answer.
Copyright: the ownership question, honestly
The uncomfortable truth vendors soft-pedal: in most jurisdictions, purely AI-generated images enjoy weak or no copyright protection, because copyright attaches to human authorship. The US Copyright Office has repeatedly declined registration for wholly machine-generated work; positions across other jurisdictions vary in reasoning but converge in practice. What this means for a seller, concretely: your licence to use the image is solid (that's contract), but your power to stop a copycat from using a very similar generated image is limited (that's copyright). Product imagery has a built-in consolation, photos anchored to your actual product carry your product's identity, and your trademarks, trade dress, and the product itself remain fully protected. The image of your ring is hard to own; the ring is not. For most commerce imagery this resolves the anxiety; for hero campaign art meant to be exclusive, it argues for meaningful human authorship in the workflow, direction, composition, retouching, which is also what strengthens protectability.
On-model imagery: the likeness layer
Worn imagery adds a third rights layer: the person in the frame. AI-generated models sidestep traditional model releases, there's no human whose likeness is used, but read the specific terms: some tools licence their model faces per-tier or restrict certain uses (endorsement-style claims, sensitive categories), and consistency features that reuse 'the same' AI model across your catalog are a licensed feature, not a right you own. One brand-safety rule regardless of contract: generated faces must not resemble identifiable real people, and any real-person photo entering try-on workflows, a customer selfie, a founder's hand, needs the same consent discipline a photographed model would get.
Resolution: the pixels side of usable
| Channel | Working minimum | Comfortable | Notes |
|---|---|---|---|
| Marketplace listings | 1,600px | 2,000 to 3,000px | Zoom features gate below ~1,600px; check each platform's spec |
| Own site / PDP | 1,200px | 2,000px | Retina displays double effective demand; weigh page speed |
| Social feeds | 1,080px | 1,350 to 1,440px | Platforms recompress everything; sharpness beats raw size |
| Paid display ads | Per placement | 2× placement size | Ad networks publish exact specs; oversupply, then crop |
| Print / packaging / OOH | 300 DPI at physical size | Vector or 4,000px+ | The tier-gate case: many tools ship print-grade only on upper plans |
Two practical notes. First, native resolution beats upscaled resolution: AI upscalers are good and getting better, but on jewellery they can hallucinate micro-detail, facet junctions, prong tips, which is exactly where the fidelity standard lives; prefer tools that generate at the size you need over tools that inflate afterwards, and re-run the zoom inspection on anything upscaled. Second, resolution gating is a common tier lever, if print or marketplace-zoom sizes only ship on higher plans, that plan is your real price.
Rights and resolution are the least glamorous part of AI imagery, and the part most worth doing once, properly, in writing. If you'd rather the vendor side of that table came pre-answered, commercial rights, survival terms, print-grade output, training opt-outs, and ask us these exact questions first; the answers are the fastest way to judge whether any imagery partner has thought past the demo.
Frequently asked questions
- Can I use AI-generated product images commercially?
- Generally yes on paid tiers of mainstream tools, whose terms grant commercial usage of outputs, but verify three things: you're on a tier that includes the grant (free tiers often exclude it), the licence to existing outputs survives cancellation, and your specific use (paid ads, packaging) isn't carved out. Contract rights, not copyright, are what authorise your day-to-day use.
- Who owns the copyright of an AI-generated image?
- Often no one, fully: most jurisdictions require human authorship for copyright, and purely machine-generated images have weak or no protection (the US Copyright Office has repeatedly declined such registrations). Your right to use the image comes from the tool's licence; your protection against copycats comes mainly from what's in the image, your product, trademarks, and trade dress remain fully yours.
- Do I need a model release for AI-generated models?
- No traditional release exists to get, there's no human whose likeness is used. Instead, the tool's terms govern how its generated faces may be used; check for per-tier licensing, restricted use cases, and whether catalog-wide model consistency is included. Real-person photos entering try-on flows still need real consent, and generated faces must not resemble identifiable people.
- What resolution do I need for AI product photos?
- Channel math: roughly 1,600 to 3,000px long edge for marketplace listings with zoom, 1,080 to 1,440px for social, 2× placement size for display ads, and 300 DPI at physical size for print. Prefer tools that generate natively at your largest requirement, upscalers can hallucinate micro-detail on jewellery, and note that print-grade output is often gated to higher pricing tiers.
- Are my uploaded product photos used to train the AI?
- Depends on the tool and your settings: many use customer content for model improvement by default with an opt-out available; enterprise tiers typically exclude training use contractually. For brands with unreleased collections or exclusive designs, confirm the opt-out in writing before uploading, your source photos are often more sensitive than the generated outputs.