Stock Photos vs AI-Generated Images: An Honest Cost and Quality Comparison

The stock-versus-generation debate is usually argued on the wrong axis. Vendors on both sides compete on price per image, which is the one dimension where the answer is obvious and the least important to the outcome.

The useful question is narrower: for this image, on this page, which source produces something a visitor believes and a crawler can understand — and what does the total work cost, including the part neither vendor does for you. As of July 2026 the honest answer splits by use case rather than declaring a winner.

Where each option genuinely wins

Stock wins when the subject must be real. A recognisable skyline, a specific building, a real event, a named product, a documentary moment with actual people. Generation cannot supply these, and attempting it produces something that is either subtly wrong or actively misleading. Stock also wins on the long tail of very specific technical subjects, where a photographer has captured something a generator has thin training coverage for.

Generation wins on specificity to your situation. This is the underrated advantage. Stock is a search problem: you look for something close to what you need and settle. Generation is a description problem: you describe the scene you want and get it. When the requirement is "a technician programming a key fob in a car footwell in overcast afternoon light, seen from the passenger side", stock will offer you eleven near-misses while generation gives you that.

Generation wins on exclusivity. Any stock image you can find, your competitor can find. The same three photographs circulate across an entire industry's websites, which is precisely why the stock look reads as low-effort. Generated imagery is unique by construction.

Generation wins on volume economics. A site needing 150 images a year is looking at a meaningful stock bill on paid libraries versus a subscription that treats that as routine. If you are still choosing a tool, our survey of AI image generators for SEO compares them on the metadata handling that matters here.

Stock wins on legal certainty for brand-critical assets. A licensed photograph has a documented chain of rights. A generated image, in the US, likely carries no copyright you can register — fine for a blog header, a problem for a logo or mascot. Our post on commercial use and copyright covers where that line falls.

The comparison in the terms that matter

Dimension	Paid stock	Free stock	AI generation
Cost per image at volume	dollars to tens of dollars	free	cents
Exclusivity	none — widely reused	none — heavily reused	unique per generation
Match to a specific brief	approximate; you settle	poor	high; you describe it
Real places, events, people	yes	limited	no
Turnaround for a new brief	minutes of searching	minutes	one to two minutes
Copyright you can register	licensed, documented	licence varies	likely none (US)
Consistency across a set	hard — different shoots	very hard	achievable via a fixed style block
Arrives with usable metadata	contributor keywords, poor filename	usually none	none
Risk of a competitor using the same file	high	very high	none

Read the last two rows together, because they are the ones that decide search outcomes and neither vendor advertises them.

The cost both options hide

A stock file downloads as istockphoto-1487293014-1024x1024.jpg with the contributor's generic keyword list attached. A generated file arrives as image.png with nothing.

In both cases someone must then decide the descriptive filename, write alt text that reflects what the image actually shows, and — if you care about provenance or local relevance — embed business, description and copyright fields. Google Search Central's image documentation names filenames and alt text among the things it uses to understand an image, so this is not optional polish; it is the layer that determines whether the file is findable at all.

Estimate it honestly. Two to four minutes per image to name it well, write real alt text, and embed metadata. At 150 images a year that is somewhere between five and ten hours of focused work — comfortably more than the licence-cost difference for most small teams, and it is work that produces no visible output, which is exactly why it gets skipped.

The observable result of skipping it is a library of istockphoto-*.jpg and image (7).png files with empty alt attributes, invisible in image search regardless of how much was paid for them. Our image SEO checklist sorts the whole list by leverage, and the filename and alt text guides cover the two items that matter most.

A worked example: a 40-page service site

Abstract cost tables are less useful than one honest count, so here is a realistic brief. A service business is building out 40 pages — 12 service pages, 20 location or area pages, 8 guides — and wants two images each. That is 80 images.

All paid stock. At mid-range single-image licence pricing this is a four-figure spend, or a subscription commitment for a burst of work that finishes in a month. The bigger cost is search time: finding 80 usable images that do not obviously come from the same three shoots is a genuinely tedious multi-day task, and the outcome is imagery your competitors can also license.

All free stock. Zero licence cost. The catch is exhaustion — the free libraries are heavily picked over, so 80 images from them will include obvious repeats and several that already appear on competitor sites in your industry. Attribution requirements vary by source and need tracking.

Generation. 80 images sits inside a normal monthly allowance on a mid-tier plan. Prompt-writing time is roughly what stock-searching time would have been, with a better hit rate because you are describing rather than hunting. Every image is unique to you.

The line neither option removes. 80 images still need 80 filenames, 80 alt texts and, if provenance matters, 80 embedded metadata blocks. At three minutes each that is four hours. This is the same in all three columns, and it is the largest single line item in the whole project.

That last point is the reason the comparison is usually framed wrongly. Licence cost varies by an order of magnitude between the options; the description work is identical and, at this scale, comparable to or larger than the licence difference. Any evaluation that ignores it is measuring the smaller number.

Neither option fixes the generic-look problem for you

Worth saying plainly, because it is the one criticism levelled at generated imagery that is equally true of stock.

The stock look — four people in a bright office laughing at a laptop — is not caused by photography. It is caused by choosing an image that could illustrate any article about any business. Generation reproduces it faithfully when the prompt is equally generic: "professional team collaboration, ultra realistic, 4K" returns exactly that composite.

The escape is the same either way: specificity. In stock, that means going past page one and preferring images with awkward, particular detail. In generation, it means naming the camera, the light direction, the setting details and the imperfections — the approach in our photorealistic prompting guide.

A practical allocation

For most small and mid-sized businesses, the sensible split is not either-or:

Photograph yourself: premises, staff, vehicles, tools, completed work, anything a customer would recognise as evidence you exist. This is your highest-value visual asset and neither option substitutes for it. A phone camera in good light beats both.

Generate: blog headers, service-page illustration, category imagery, abstract and background plates, scene-setting shots where the subject is a situation rather than a specific real thing. This is the bulk of most sites' image needs by count.

Licence stock: the specific cases generation cannot serve — real places, real events, identifiable products, editorial subjects — and anything brand-critical where you want documented rights.

Never generate: your premises, your staff, your completed work, your products. Presenting a generated image as documentary evidence of your business is a trust problem long before it is a legal one.

The licence risks worth knowing on both sides

Neither option is risk-free, and the risks are different enough to be worth naming.

On the stock side, the recurring problem is scope rather than the image itself. A standard licence typically covers use on your website and marketing materials but not merchandise, not logos or trademarks, not redistribution, and not use as the primary element of a product for sale. Editorial-only images are restricted to news and commentary contexts and may not be used in advertising at all — a distinction that catches people out because the images look ordinary. And a subscription lapse does not always leave you with perpetual rights to what you downloaded, which is worth reading rather than assuming.

On the generation side, the risks are the two covered in our copyright post: limited or absent registrable copyright in the raw output, and content-level exposure if an image contains a recognisable trademark, a likeness resembling a real person, or something substantially similar to an existing protected work. Prompting for named artists, characters or brands is where that exposure comes from, and avoiding those three categories removes nearly all of it.

The common thread is record-keeping. Whichever source you use, note where each asset came from, under what licence or plan, and when. That file takes minutes to maintain and answers every awkward question later — from a client, an auditor, or a rights holder — in seconds.

Where the metadata work lands

Under that allocation the generated bucket is the largest by volume, which means it is also where the per-file metadata cost concentrates. This is the specific gap SEOpix addresses: it writes the descriptive filename, alt text and EXIF block — business name, description, keywords, copyright, optional GPS coordinates — at generation time, so files enter your library described rather than joining a backlog.

That does not make generated imagery better than stock. It makes the metadata cost, which is the real cost in both columns, disappear from one of them. For teams working across many locations or clients, the bulk workflow guide covers how that plays out at a few hundred images a month, and geo-tagging covers the local-relevance layer.

Pick per image, not per philosophy. Then make sure whichever file you end up with has a name and a description, because that is the part that decides whether anyone finds it. Ten images a month are free if you want to compare the workflows directly, and the pricing page has the volume tiers.