FOTOhub at scale: the technology, the offer and the full price list behind 1,000,000 users

One year ago today, on September 11, 2025, FOTOhub.app shipped an AI image generator. Twelve months later the same company runs 226 AI models for more than 1,000,000 users in 46 countries, serves 314 public API endpoints, trains and hosts its own generative models on its own GPU fleet, and sells cloud compute by the hour. It ranks #265 in the world on Crunchbase CB Rank, and its founder and CTO Mateusz Ulewicz is ranked #1 globally among more than 2.1 million tracked profiles. DGE3 Investment led its Seed round in February 2026 and has committed $10.8 million to date.

We have written about the milestones. This post is the engineering and commercial audit underneath them: how the platform actually scales, what it now sells, and exactly what it charges. If you are evaluating FOTOhub as a user, a developer, an acquirer or a co-investor, this is the document we would hand you.

Sourcing note, because a fund publishing numbers about its own portfolio company should say where they come from. Product and infrastructure figures are taken from FOTOhub's engineering posts of August and September 2026 and from docs.fotohub.app; pricing is the plan matrix published at fotohub.app/public_prices as of September 2026; credit costs come from the SDK reference; capital figures are DGE3's own committed position. Where the company's own posts disagree with each other — and on a platform shipping this fast, they sometimes do — we use the most recent figure and say so. Prices and model counts move; verify before you buy.

The scaling curve, quarter by quarter

Growth is easy to assert and hard to compound. Here is the actual curve, with the platform surface that had to be built to carry it.

Two observations we would make about that table if it belonged to a company we were meeting for the first time.

First, the model count stops being the story after August. It went from 163 to 226 in six weeks and then stopped moving, while endpoints, studios and integrations kept climbing. That is the signature of a company that finished its breadth phase and moved to depth — exactly the sequencing you want, and the opposite of what most catalogue businesses do.

Second, the user curve is not campaign-shaped. Paid acquisition has stayed below 10% of signups through the whole period. A funded competitor can buy 500,000 signups; it cannot buy 62% week-eight retention, which is what FOTOhub reports.

The external scoreboard agrees. On Crunchbase, the industry's reference database for private-company momentum, FOTOhub ranks #265 in the world on CB Rank (Organization) as of September 2026, and its founder and CTO Mateusz Ulewicz holds the #1 position globally on CB Rank (Person) — first place out of more than 2.1 million tracked profiles. We have backed companies that grew faster in a single quarter. We have never backed one that reached the global top 300 of that ranking inside twelve months of launch.

The orchestration layer: why 226 models behave like one

The single most valuable asset in this company is invisible to its users. Every model provider has its own request format, its own error taxonomy, its own rate limits and its own habit of deprecating endpoints on short notice. FOTOhub absorbs all of it in one place, and everything else it sells depends on that.

What the layer does, concretely:

• Health monitoring and provider isolation. A degraded model is pulled from rotation automatically and reports itself unavailable rather than accepting work it cannot finish. One provider's bad hour is not FOTOhub's bad hour.
• One job model across every medium. Image, video, audio, 3D and chat jobs share a single asynchronous job contract with one job id that the app, the REST API and the MCP server all understand.
• Streaming everywhere. First-token streaming on every provider that supports it, including through the public API.
• Price before execution. Every operation on every surface quotes its credit cost before it runs. This sounds like a billing detail; it is the reason enterprise buyers take the platform seriously.
• Gabriel, the router. Ask for a result rather than a model and Gabriel selects the model, sets parameters and routes the job. It runs on FOTOhub's own GPUs, which is why it can also serve as the free tier's chat model at no marginal token cost — a neat piece of cost engineering that doubles as a funnel.

The commercial consequence: a new frontier model reaches FOTOhub users within days of its public release, at the same price list as everything else, with no renegotiation of the platform's cost structure. Seedream 5.0 PRO, IDA Q 1.0, Seedance 2.5 and Luma Ray 3.2 all landed inside eight weeks.

Owning the stack: the IDA family

In July 2026 FOTOhub stopped being purely a distributor of other companies' models. IDA Q 1.0, its first proprietary image model, was trained and is served on the company's own GPU fleet. The company reports it in the global top five on the DesignArena benchmark and roughly 20x cheaper to run than GPT-Image-2 — its number, not ours, but the economics are visible in what the platform can afford to give away.

The rest of the owned stack:

• IDA Voice — the speech engine behind narration across the studios and the text-to-speech tool exposed to agents.
• IDA Y — music, sound effects and soundtrack generation.
• Gabriel — orchestration and the free-tier chat model.
• The tooling layer — upscaling, face restoration, depth maps, inpainting and outpainting, background removal, plus two tiers of 3D generation, all on own hardware.

Why this matters more than any single launch: owning the model that handles the baseline load puts a floor under gross margin. It is what makes "unlimited Seedream on Professional and Business" a sustainable offer rather than a promotion, and it converts every future third-party price increase from an emergency into a routing decision. In our February underwriting we did not model vertical integration at all. It is now the variable we track most closely.

The infrastructure: AWS, an own GPU fleet, and selling the surplus

FOTOhub completed a full production consolidation onto Amazon Web Services in August 2026 and is a member of the AWS Startup Portfolio Tier with a paid AWS Business Support agreement. AWS backed the company with $100,000 in Activate credits in March 2026 and announced a joint global scaling plan in June.

The strategic point is the last row of the product, not the infrastructure: the same fleet that renders FOTOhub's own jobs is rented to customers by the hour. A cost centre became a revenue line, and regional data residency stopped being a roadmap item and became a feature a regulated buyer can put in a contract. Very few companies at this stage own hardware; almost none monetize the slack.

The developer platform: the second business

FOTOhub's second customer is a developer, and the second half of year one was largely built for them. Our February memo treated the API as 2027 optionality. It shipped in July and is now a platform in its own right.

Two details in that table are worth more than they look. Being both an MCP server and an MCP client means FOTOhub can be the place work gets done and a tool other agents call — very few companies sit on both sides of that protocol, and it is a distribution channel that did not exist as a category when we underwrote the Seed. And the SDK release cadence — ten versions in 23 days — is the clearest public evidence of the engineering rhythm we describe in our portfolio reporting.

The offer: thirteen studios, one balance

Models are components. Studios are what a customer opens. All thirteen draw on the same catalogue, the same credits and the same asset library.

The catalogue those studios draw on, by category: 63 image models (Seedream 5.0 and 4.5, GPT Image, Imagen, IDA Q 1.0), 65 video models (Seedance 2.5, Luma Ray 3.2, Veo, Sora), 30 audio operations (text to speech, music, sound effects, stem separation, mastering, transcription, consent-gated voice cloning), 35 chat models (GPT, Claude, Gemini, DeepSeek, Nova families), 20 text models, plus 3D generation, virtual try-on and utility models. The interface ships in 25 languages.

The price list, in full

This is the part most analyses skip. Below is the plan matrix as published on fotohub.app/public_prices in September 2026, with the effective cost per credit that we calculate from it. Annual billing is roughly 17% cheaper than the monthly rates shown.

And the one-time top-ups, which stack on any plan without an upgrade and lift the four-hour burst limit:

Buster Packs are billed in Polish złoty regardless of the currency chosen for the subscription — a detail worth knowing if you are reconciling card statements in another currency.

What the pricing structure tells an investor

Three things, and they are the reason we consider the price list a strategic document rather than a marketing page.

The curve is smooth and the top-ups are honest. Cost per credit falls from $0.0267 on Starter to $0.0066 on Team — a 4x volume discount — and every Buster Pack is priced close to the plan tier a customer would have upgraded into. Running out of credits mid-month is not punished. Platforms that penalise overage train their users to under-use them, which is the single fastest way to kill a usage-based business.

Unlimited on the most-used models is only possible because of IDA. Giving Professional and Business subscribers uncapped Seedream 5.0, Seedream 4.5 and GPT Image 1.5 inside a daily fair-use ceiling would be reckless for a pure reseller. It is defensible for a company whose own model anchors the cost of the most common operation on the platform.

One balance covers the developer surface. No separate API plan, no MCP surcharge, the same per-operation prices. Commercially this is a land-and-expand mechanism: the $22 consumer who writes a script becomes the $175 account, and then the account that buys GPU hours. It is also the cleanest explanation we have for net revenue retention above 145%.

What each plan actually buys

Credits are abstract, so here is the conversion. The cheapest image generation costs 1 credit. Video is priced per second of output: roughly 6.4 credits per second at 480p, 14.5 at 720p, and 17.6 at 720p when a video reference is supplied. Music costs 5 credits up to 30 seconds, 10 up to 60 seconds, 25 beyond that, capped at 300 seconds. A virtual try-on render is 4 credits for a single garment, 3 for a chained top-and-bottom job.

The "off-balance" column is the one that changes buying behaviour. A Business subscriber generating on Seedream 5.0 within the 250-images-per-day ceiling can produce 7,500 images a month without touching the credit balance, leaving all 20,000 credits for video, audio and 3D. At $175 a month, that is a production capacity most agencies were paying five separate vendors for a year ago.

The stack it replaces

FOTOhub's own pricing analysis benchmarks the platform against the subscriptions a working creator typically stacks. We reproduce the comparison because it is the clearest statement of the wedge — with the caveat that these are list prices as published in FOTOhub's September 2026 analysis, and competitor pricing changes frequently. Verify current rates before making a purchasing decision.

The comparison understates the difference in one respect that no price table captures. Ten subscriptions do not merely cost more; they do not compose. An image made in one tool has to be exported, converted and re-uploaded to be animated in another. On FOTOhub the output of any model is the input of any other, in one library, on one balance, with one invoice.

Built and hosted in Europe — and what is still missing

This morning FOTOhub published its position on data residency, which we read as the opening move of its enterprise phase. The substance: the company is a Polish operating company headquartered in Bydgoszcz with a Delaware entity for the US market; its proprietary models — audio, speech, lip sync, the IDA image family, the tooling layer and 3D generation — run on FOTOhub's own hardware in Frankfurt and Ireland and that data does not leave its infrastructure; third-party partner models that process data outside Europe are named individually in the privacy policy.

We are publishing the last two rows deliberately. SOC 2 Type II and ISO 27001 are the next gate for the enterprise and public-sector contracts this platform is otherwise ready to serve, and content provenance is becoming a procurement question rather than an ethics debate. A company that states plainly what it does not yet hold is a company whose other claims we are inclined to believe — and both items are on the agenda for the capital we are raising.

Where the numbers disagree

Anyone who reads FOTOhub's blog end to end will find figures that do not reconcile, and we would rather explain them than let a reader conclude someone is inflating.

• Models: 163 or 226? The August 1 post counts 163 catalogued with 110+ production-enabled; from August 18 the figure is 226 across eight categories. Both are true at their date. Treat 226 as the current catalogue, not an audited constant.
• Regions: 5 or 16? The earlier figure counts production regions running compute; the current one counts AWS regions available as storage targets for customer output. Different measures, both accurate.
• Capital: $8.6M or $12.3M? FOTOhub's public figure predates the completion of our August expansion tranche in its press materials. Our committed position is $10.8 million of an approximate $12.3 million total. We set out the full round-by-round ledger in our first-anniversary review.
• Crunchbase. As of September 2026, founder Mateusz Ulewicz holds the #1 position worldwide on the Crunchbase global CB Rank (Person) leaderboard — first out of more than 2.1 million tracked profiles — and FOTOhub itself ranks #265 globally on CB Rank (Organization). Older DGE3 posts cite top-ten and top-five positions; those were accurate at their date, on a ranking that updates continuously. The company figure is the one to watch: a twelve-month-old private company inside the global top 300 of a database tracking millions of organisations is an outlier by any measure.

Our read, and what we are funding next

A year in, FOTOhub is not a creative tool that grew fast. It is an AI infrastructure company with consumer distribution: an orchestration layer that makes 226 models behave as one, proprietary models carrying the baseline load, a GPU fleet whose surplus is a product, and a developer platform that arrived four quarters ahead of our plan. Roughly $12 of capital per user, at a $180 million Q2 2026 valuation, on approximately $12.3 million raised in total.

The four bets in our year-two model, in descending order of conviction: more platform load on owned IDA models; agents that execute complete workflows rather than single tools; developer and enterprise revenue as a growing share of the mix; and distribution through protocol — MCP, plugins, marketplaces — rather than media spend. Add to that the two certifications above, and the picture of what the next round funds is complete.

DGE3 is assembling a $100 million Series A syndicate for FOTOhub: model training capacity, regional infrastructure, the developer platform and a United States go-to-market build. We are in conversation with growth funds, strategic investors and sovereign technology platforms. If that is your mandate, write to invest@vcdge3.com.

Frequently asked questions

How much does FOTOhub cost? Free with 50 credits and no card; Starter $8 (33 zł) for 300 credits; Medium $22 (89 zł) for 1,500; Professional $65 (260 zł) for 5,000; Business $175 (700 zł) for 20,000; Team $99 per seat (386 zł) for 15,000 credits per seat, 3–50 seats. Annual billing is about 17% cheaper. Professional and Business also generate on Seedream 5.0, Seedream 4.5 and GPT Image 1.5 without spending credits, within daily fair-use ceilings of 60 and 250 images.

What are Buster Packs? One-time credit packs that stack on any plan and lift the four-hour burst limit: 200 credits for $3.99 (7 days), 500 for $7.99 (30 days), 1,500 for $17.99 (30 days), 5,000 for $49.99 (no expiry), 15,000 for $124.99 (90 days), 50,000 for $374.99 (no expiry). Billed in Polish złoty regardless of subscription currency.

Is the API included in the subscription? Yes — 314 REST endpoints, Python, TypeScript and PHP SDKs, a CLI, an n8n node and a 28-tool MCP server, all on the same balance at the same prices. No separate developer plan, no MCP surcharge. Default 60 requests per minute per key, adjustable to 600.

How many AI models does FOTOhub have? 226 across eight categories as of September 2026: 63 image, 65 video, 30 audio operations, 35 chat, 20 text, plus 3D, virtual try-on and utility models, including FOTOhub's own IDA Q 1.0.

Does FOTOhub have its own AI models? Yes: IDA Q 1.0 for image, IDA Voice for speech, IDA Y for music and sound, the Gabriel orchestrator, and the tooling layer — all trained and served on FOTOhub's own GPU fleet in Frankfurt and Ireland.

Where is FOTOhub hosted? On Amazon Web Services, with its own GPU fleet in Frankfurt and Ireland and storage selectable across 16 AWS regions. It is a Polish company with a Delaware entity for the US market, and it does not yet hold SOC 2 or ISO 27001.

How fast is FOTOhub growing? 460,000 users in February 2026, 600,000 in April, 750,000 in June, 926,000 in mid-August and more than 1,000,000 in 46 countries in September 2026, with paid acquisition below 10% of signups, 9.5% paid conversion, 145%+ net revenue retention and 62% week-eight retention.

What is FOTOhub's Crunchbase ranking? As of September 2026 FOTOhub ranks #265 globally on Crunchbase CB Rank (Organization), and founder and CTO Mateusz Ulewicz holds the #1 position worldwide on CB Rank (Person), first out of more than 2.1 million tracked profiles.

Can I invest in FOTOhub? DGE3 Investment is assembling a $100 million Series A syndicate. Write to invest@vcdge3.com; company materials are at fotohub.app/investor-relations.

Further reading

• FOTOhub turns one: a DGE3 investor review of twelve months, 1 million users and 226 AI models on AWS
• FOTOhub crosses 1 million users: one plan replaces ten AI subscriptions
• DGE3 opens FOTOhub's expansion tranche — total commitment $10.8M
• FOTOhub by the numbers: 226 models, AWS consolidation, new COO
• The creative AI market in 2026 and why DGE3 is all-in on FOTOhub
• FOTOhub company profile in the DGE3 portfolio
• FOTOhub plans and prices · developer documentation · Creative AI built and hosted in Europe

DGE3 Investment is a multi-strategy venture capital fund investing in fintech, AI, digital assets, deep tech and real assets from Warsaw, Prague and Abu Dhabi. FOTOhub is a DGE3 portfolio company; this post is published for information purposes and is not an offer to sell or a solicitation to buy securities, nor a recommendation to purchase any product. Prices, model counts and platform figures are as published by FOTOhub in September 2026 and change frequently; verify at the source before relying on them. Portfolio marks are as reported at the most recent priced round.

#FOTOhub #FOTOhubPricing #CreativeAI #GenerativeAI #AIOrchestration #AIInfrastructure #AWS #AmazonWebServices #MCP #IDAQ #Gabriel #Seedream #Seedance #DeveloperPlatform #APIFirst #VentureCapital #SeriesA #DGE3 #Crunchbase #MateuszUlewicz #PolishTech #EuropeanAI #DataResidency #ScalingUp