60% of Bestselling "Photos" Aren't Photos at All
Spoiler: The picture topping your favorite stock site last week never saw a lens.
The Week We Measured
Every June brings a rush of mid-year marketing briefs. Brands need fresh creative, and art buyers flock to microstock portals for fast answers. To see how far artificial imagery has crept into that pipeline, we scraped Adobe Stock’s daily “top sellers” feed for 23–29 June 2025.
From that pile we filtered down to items whose media_type_label reads “Photo.” One hundred twenty records met the rule. Each entry carries a silent switch called is_gentech. A true value means the file was born in a generative model rather than in a camera. That switch became the hinge of our study.
[Image: prompt a 3d low-poly collage of vintage cameras melting into neural network nodes, pastel palette, soft global light, no text, no numbers, center composition]
Checking the Flag
Adobe keeps the definition plain. If Firefly or any rival network created the pixels, the asset counts as Generative AI. If a sensor captured the photons, the flag stays false. The field is not perfect — creators can mis-label or forget the declaration — but it gives the clearest, public signal we have today. Because these files invite broad commercial use, mis-labeling risks account suspension, so most contributors keep the stamp honest.
The Core Count
We wrote a three-line script, tallied every flag, and found a neat split:
+-------------------+-------+-------+----------+
| Scope | Real | AI | AI Share |
|-------------------+-------+-------+----------|
| Top 120 photos | 48 | 72 | 60% |
+-------------------+-------+-------+----------+Seventy-two photos outperformed every camera shot that week. Six in ten “photographs” began as tokens, not glass. The ratio alone feels dramatic, yet the details show an even sharper story.
Where AI Wins at 100 Percent
We grouped each record by the category tag that Adobe adds for fast search. Ten buckets held at least two files, so they formed our next table:
+----------------------------+-------+----+------+--------+
|Category |Total |AI |Real |AI% |
+----------------------------+-------+----+------+--------+
|Food |15 |15 |0 |100.0% |
+----------------------------+-------+----+------+--------+
|Lifestyle |10 |10 |0 |100.0% |
+----------------------------+-------+----+------+--------+
|Landscapes |9 |9 |0 |100.0% |
+----------------------------+-------+----+------+--------+
|Drinks |7 |7 |0 |100.0% |
+----------------------------+-------+----+------+--------+
|Industry |2 |2 |0 |100.0% |
+----------------------------+-------+----+------+--------+
|Buildings and Architecture |2 |2 |0 |100.0% |
+----------------------------+-------+----+------+--------+
|Science |2 |2 |0 |100.0% |
+----------------------------+-------+----+------+--------+
|Plants and Flowers |2 |2 |0 |100.0% |
+----------------------------+-------+----+------+--------+
|Travel |2 |2 |0 |100.0% |
+----------------------------+-------+----+------+--------+
|Culture and Religion |2 |2 |0 |100.0% |
+----------------------------+-------+----+------+--------+Food shots led every download chart, and every plate, bottle, or garnish was synthetic. Busy art directors no longer wait for a kitchen studio. They type “sun-lit ramen with steam” and get a perfect bowl in thirty seconds. Lifestyle images show the same trend. Models never blink, release forms arrive by script, and costs drop to zero.
The Human Holdouts
Four segments still lean on real pixels:
+-------------------+-------+----+------+-------+
|Category |Total |AI |Real |AI% |
+-------------------+-------+----+------+-------+
|Business |9 |3 |6 |33.3% |
+-------------------+-------+----+------+-------+
|Graphic Resources |26 |6 |20 |23.1% |
+-------------------+-------+----+------+-------+
|People |14 |1 |13 |7.1% |
+-------------------+-------+----+------+-------+
|Technology |5 |0 |5 |0.0% |
+-------------------+-------+----+------+-------+Why do people and tech still hug the lens? Buyers trust old-fashioned skin in close-ups, and circuit boards remain tricky for diffusion engines. Even so, the presence of any AI in “People” signals a shift. Projects that once needed five crew members now finish with a prompt and a tweak.
The Method in Plain Words
- Query the Feed: We hit Adobe’s public endpoint each morning at 09:00 UTC.
- Store the JSON: We saved every entry in a database, keeping the ID, title, category, and is_gentech value.
- Filter by Media Type: We removed illustrations and vectors to focus on photographic intent.
- Count the Flags: A tiny function walked the list and produced the tables above.
- Spot-check Thumbnails: We opened random AI items to confirm that obvious tells — plastic light, impossible glass refraction — matched the flag.
The run took under two minutes on a mid-range laptop. The hardest part was waiting seven days so the sample covered a full work week.
The Big Question and a Direct Answer
Why does this shift matter to a working stock photographer? Because the same buyers who once licensed your macro of dewy basil now pick a friction-free clone that costs less than a coffee. If you shoot stock full-time, the rules changed while you slept. No trend in the last decade moved the goalposts this far, this fast.
Reading Between the Pixels
Three signs point to a deeper change:
- Speed beats craft. Art teams value instant turnarounds for social content that lives only twenty-four hours.
- Cost drives selection. AI images enter most libraries at the lowest standard price. Real photos carry production overhead.
- Control wins briefs. Generative tools let designers nudge colors, swap props, or widen aspect ratios without a reshoot.
Traditional shoots still shine in jobs that need authentic faces, real gear, or strict legal clarity. Yet the comfort zone narrows.
Show Me the Money
Generative art shook the
leaderboard, but it also shifted paydays. Adobe pays a flat 33 %
royalty on every non-video license. Subscription plans bring the download fee down to about
US $0.33–0.38 for heavy sellers.
That range turns
small counts into coffee money and large counts into rent.
We matched each top-seller with its creator name, then asked a short question: How does the cash divide between camera users and prompt writers? The answer rests in a quick projection. Each license below assumes the conservative US $0.35 payout.
+-----------------------+---------+-----------+---------------+
|Creator |Images |AI Share |Projected $/wk |
+-----------------------+---------+-----------+---------------+
|sri |12 |100% |$147.0 |
|regina |12 |100% |$147.0 |
|ferdian |12 |100% |$147.0 |
|Minol |12 |100% |$147.0 |
|Ani |12 |100% |$147.0 |
|arista |12 |100% |$147.0 |
|Deemerwha studio |12 |0% |$147.0 |
|chathuporn |12 |0% |$147.0 |
|fatima |12 |0% |$147.0 |
|Stock 4 You |12 |0% |$147.0 |
+-----------------------+---------+-----------+---------------+The projection multiplies 12 images × 35 downloads × $0.35. We picked 35 because the lowest rank crossed that mark on three random refreshes. The pattern holds even if real photos win a few extra downloads; AI shoots scale faster because one person can publish hundreds a week.
Keyword Wars
A download begins at the search box, so we mapped the first twenty keywords from every title tag. Two clear camps appear.
+-------------------+-------------------------+----------------------------+
|Rank |AI Top Terms |Real Photo Top Terms |
+-------------------+-------------------------+----------------------------+
|1 |texture (47) |business (24) |
|2 |design (41) |design (20) |
|3 |simple (38) |technology (20) |
|4 |light (35) |office (19) |
|5 |background (32) |isolated (17) |
|6 |clean (30) |concept (17) |
|7 |white (29) |corporate(17) |
|8 |natural (28) |element (16) |
|9 |modern (27) |background(16) |
|10 |minimal (26) |success (15) |
+-------------------+-------------------------+----------------------------+AI artists lean on texture, simple, and clean because those prompts yield flat colors and no artifacts. Traditional shooters load briefs with business, office, and corporate to sell serious slides. The overlap on design hints at an arms race: the word stays broad enough to hit any trend, so both camps spam it.
What does that mean for you if you still aim a camera? Aim at nouns that diffusion misses — wrinkles, rust, messy desks, and other honest clutter. Fill metadata with honest detail, then let the algorithm sort.
Emerging Playbook for Contributors
- Batch niche prompts. Publish entire series around one missing subject, such as left-handed tools or tropical houseplants in office light.
- List hard facts. Add explicit numbers, model years, or ingredient weights to crush broad spam.
- Sync seasons early. Upload holiday sets four months ahead; search volume spikes long before ad budgets lock.
- Watch the flag. Always tag AI work, because Adobe audits mis-marks and freezes payouts.
Answering One Big Question
Will AI push real stock photography to zero? No. Brands still crave “real faces, real places” for banner work, and no diffusion model nails that level of likeness under current license law. Yet the mid-tier catalog — textures, food plates, staged lifestyle — now lives in prompt land. Active photographers who relied on those safe staples must pivot to deeper stories fast.
Countdown to the Next Update
We will rerun the scrape in September after back-to-school campaigns settle. Expect the AI share to rise above 70 % if category patterns hold. We showed the first warning and gave the numbers and the money trail. The final section will drop in that fall report and trace whether royalty rates stay flat or tilt when subscription prices shake.
Keep your shutter ready, but keep your prompts sharper.