AI Generation × Adobe Stock: ¥500,000 a Year. Twelve Months of Real Data Behind the ¥42,000/Month a Salaryman in His 40s Built on Weekends Alone

The published numbers (2025)

Side-income disclosures tend to cherry-pick the “best month,” but this is a rare case where all twelve months of actual figures are laid out. That makes it possible to examine what sits inside the ¥42,000 monthly average, month-to-month variance, seasonality, and three years of accumulation.

What the business is

Rossin is a company employee in his 40s who creates assets with AI image-generation tools and registers and sells them on Adobe Stock. He started in 2022 and works weekends only, about 2–3 hours per session. He ranks around 17,900th in Adobe Stock’s creator rankings, and in addition to Adobe Stock as his main platform he has expanded to PIXTA. Demand for stock assets tracks the working calendar of design practice, assets for the New Year holidays and seasonal events create the sales peaks. From the twelve months of published actuals, you can see that stock-type income moves on “accumulated volume of uploads × seasonal demand.”

Reading the monthly data

Lining up the twelve months reveals several structural patterns.

• February’s ¥28,000 and December’s ¥56,000 are exactly a 2x spread. December, when demand concentrates on New Year’s cards, Christmas, and the year-end shopping season, is the peak; February, the demand trough, is the minimum
• The first half (January–June) totals ¥223,000 versus ¥279,000 for the second half — about 25% more. The density of events from autumn onward feeds directly into sales
• Second only to December is September at ¥49,500. By his own breakdown, cherry-blossom assets peak in March–April, summer assets in July–August, autumn assets in September–October, and Christmas/New Year assets in December
• The interesting dip is November’s ¥39,000. Sales that climbed through September–October fall back in November, then jump to the annual high in December. The movement suggests that the working calendar of the buyers (designers and production companies) splits into two waves: the end of autumn production and the year-end rush
• Note too the “floor”: even the worst month, February, brought in ¥28,000. If the income depended on a few hit assets or a search trend, off months would approach zero. The natural reading is that the monthly floor holds because three years of registered inventory sells thinly across a wide range of uses

This seasonality translates directly into a playbook variable. The factor he names first among his earnings drivers is producing seasonal assets 3–6 months ahead of demand, assets that sell in December are prepared between summer and autumn. With the same production volume, registration timing changes the sales.

The reasons he says he earned

Rossin himself organizes his 2025 earnings drivers into three: producing seasonal content 3–6 months in advance, continually expanding his portfolio (number of registered assets), and improving prompt quality rather than relying on volume. The third is an interesting sign of operational maturity. The practice of AI generation has become well-worn: trial and error with prompts keeps driving down the generation cost per asset, and patterns for handling rejections (Adobe Stock’s quality standards) accumulate. The meaning of three years of continuity lies not only in the inventory of assets but in the accumulated learning of “what passes review and what sells.”

Where does the 10x gap with the traditional model come from?

Within the same stock-photo world, real data from the traditional (shooting-based) model shows a cumulative ¥185,000 over four years, roughly ¥3,900 per month when averaged out. Against Rossin’s monthly average of ¥41,819, that is about a 10x gap. Dismissing this gap with “because it’s AI” misses the structure. Broken down: (1) unit production cost, shooting carries high per-asset costs due to equipment, travel, and subject constraints, while AI generation mass-produces within a subscription fee, (2) freedom of supply. The traditional model can only offer what can be photographed, whereas the generative model can build assets backward from “demand that sells”, (3) seasonal fit, preparing targeted assets 3–6 months ahead of demand is hard when shooting is bound by weather and seasons. What the 10x gap really measures is not image quality but whether supply can be designed starting from demand.

Our take

AI image generation has changed the barrier to entry in stock photography from “shooting ability” to “demand-selection ability.” Now that camera gear and photo trips are unnecessary, the only thing separating earners is the eye for “which demand (season, use case) to supply assets to.” The change in production cost structure translates directly into the earnings gap.

¥40,000 a month is a realistic destination for “weekends-only compounding.” It is not a one-off hit but a structure in which three years of accumulated uploads sell steadily every month. The fact that monthly variance stays within the ¥28,000–56,000 range is itself evidence that the income source is spread across the whole portfolio rather than concentrated in a few hit assets. The biggest risk, platform policy changes on AI-generated assets, has the same structure as Kindle publishing.

Conditions for replicating this case, and its limits

The conditions for replication are clear: (1) being able to keep paying the subscription fees for Midjourney and the like (a few thousand yen per month), (2) sustaining a few weekend hours for three years, (3) working backward from what sells (use case, season) and generating and registering assets 3–6 months ahead of demand. With shooting equipment, models, and location costs reduced to zero, the initial investment is a small fraction of traditional stock photography. The scale is designed to run on a salaried employee’s disposable time, which also makes it easy for Japanese side-business readers to test.

On the other hand, the biggest risk cannot be eliminated by one’s own efforts. Platform policies on AI assets (whether they can be registered, labeling obligations, royalty rates) can change, and price erosion from the oversupply of AI assets continues. A ranking of 17,900th earning ¥42,000 a month also means, read the other way, that this is a market with an enormous number of competitors above. “Today’s ¥40,000 a month” is a number with no guarantee of permanence. The realistic operating approach is to spread risk across multiple platforms, as he himself does by also selling on PIXTA, or to combine with the traditional model.

Related cases

• Four years of real data from traditional stock photography — the contrasting case
• 24 Kindle books, ¥50,000–70,000/month — the same “build the shelf” model