FOTOhub ships its most stable release yet, with 226 AI models under one orchestration layer

Summary

FOTOhub has released the most stable version of its platform to date, and with it a set of new tools that change what a single creative account is able to reach. The headline number is 226 AI models, available to one user, under one balance, inside one interface, spanning image, video, audio, chat, text, utility, virtual try-on, and 3D. The more interesting number is the one underneath it: 926,000 registered users across 46 countries, a base that has grown to within striking distance of a million people without the platform ever asking any of them to choose a vendor, install a client, or manage a key. This release is about consolidation rather than novelty. Everything that was previously a promising surface, the generator, the video studio, the live chat workspace, the canvas board, the automated flows, and the vertical video factory, is now production hardened, connected to the same billing and the same model registry, and shipped as a single product rather than a family of adjacent experiments. Alongside that, FOTOhub is now part of the Amazon Web Services partner portfolio, with its developer console integrated directly into AWS storage, cloud computing, and GPU capacity across sixteen regions worldwide. The platform now serves three clearly separated segments, consumer creators, business teams, and developers building on the public API and the e-commerce bridge documented at docs.fotohub.app.

What actually shipped, and why stability is the story

There is a particular failure mode in AI products that nobody puts in a press release. A platform integrates a dozen providers, each with its own request shape, its own error grammar, its own idea of what a rate limit is, and its own habit of deprecating an endpoint with two weeks of notice. The demo works. The catalogue looks enormous. Then a customer sends a hundred real jobs through it on a Tuesday afternoon and discovers that a third of them silently failed, a handful were charged twice, one provider returned a success code over an empty payload, and the queue has no memory of any of it.

The release FOTOhub is shipping this month is the answer to that failure mode rather than a new coat of paint over it. Every generation on the platform, whether it originates from a click in the dashboard, a message in the chat workspace, a node in an automated flow, or a call from a customer's own server, now travels the same path: one registry that knows what each model costs and what it can accept, one accounting layer that reserves before it spends and refunds when a provider fails, one job store that survives a restart, and one delivery contract that hands back a usable asset rather than a provider-shaped envelope the caller has to learn to unwrap. Provider quirks are absorbed at the edge, where they belong, instead of leaking into the interface where a creator would have to care about them.

That is the unglamorous work that makes a catalogue of 226 models mean something. A long model list is trivially easy to publish and extraordinarily difficult to keep honest. FOTOhub's list is the billable catalogue, which is to say the set of models a customer can actually call today at a price the platform will actually charge, distributed across roughly sixty-five video models, sixty-one image models, thirty-five conversational models, thirty audio models, twenty text models, twelve utility operations, three virtual try-on models, and three 3D models. Health monitoring runs continuously underneath that list, and when a provider degrades, the affected model reports itself as unavailable rather than accepting work it cannot finish. A creator who has never heard the word orchestration nevertheless experiences the result of it, which is that the thing they clicked returns the thing they wanted.

The generator, rebuilt as a single surface for sixty-one image models

The image generator at fotohub.app/generate/new is the clearest demonstration of what a shared orchestration layer buys a user. On the surface it is one prompt box, one model picker, and a results rail. Underneath, that single surface fronts every image model on the platform, from fast drafting models that return in a few seconds to premium models that hold a face consistent across a whole series, and it presents all of them through one identical set of controls.

Aspect ratio, batch size, seed control, negative prompting, and reference imagery behave the same way regardless of which model is selected, and where a particular model genuinely cannot honor a control, the interface says so rather than accepting the parameter and quietly discarding it. Reference-driven generation is the part experienced users notice first, because it is the difference between a model that produces a beautiful stranger and a model that produces the person, product, or brand asset you actually need. Upscaling, background removal, object removal, and mask-based editing sit in the same workspace instead of in a separate tool with a separate history, which means an image can be generated, corrected, cleaned, and enlarged without ever leaving the page or losing provenance.

Around that core the platform ships more than a hundred purpose-built studio applications, each of which is the same engine with the expertise pre-loaded: headshots and avatars, product photography, logo and brand marks, thumbnails, tattoo design, restoration of damaged family photographs, interiors, figures and collectibles, three-dimensional print preparation, and virtual try-on for apparel. This matters commercially more than it matters technically. A small business owner does not want a model picker. They want a photograph of their product on a marble surface in daylight, and they want it in one attempt. The studio applications are how a general-purpose generation stack gets delivered to somebody who has no interest in prompt engineering, while the raw generator remains fully exposed for the people who do.

Video AI: sixty-five models and the operations that surround them

The video workspace at fotohub.app/generate/videoai is the largest single surface FOTOhub has ever shipped, and the one where orchestration stops being an architectural preference and becomes the entire product. Video models disagree with each other about almost everything. Some accept a still image as a starting frame and some do not. Some accept a first and last frame and interpolate between them. Some generate their own soundtrack and some return silence. Some are priced per second, some per clip, some per resolution tier. Some return a finished file and some return a job identifier that has to be polled for several minutes. A platform that exposes sixty-five of these through one interface has to either normalize all of that or push it onto the customer, and pushing it onto the customer is how video tools end up with a support queue longer than their feature list.

FOTOhub normalizes it. One workspace covers text-to-video and image-to-video, reference-driven generation for character and product consistency, clip extension for stretching a short result into a longer sequence, frame interpolation, native audio where the model supports it, and post-generation operations including colorization, background removal, object removal, inpainting, face restoration, and lip synchronization. Duration, resolution, aspect ratio, and motion intensity are presented as one consistent set of choices, and the price is shown before the job is submitted rather than discovered afterward on a statement. Long jobs return immediately with a tracked identifier and finish in the background, so a browser tab closing does not cost anyone a render.

The practical effect is that a creator can plan a video around what they want to say instead of around which vendor they happen to have an account with. A product clip can start life as a still photograph from the image generator, gain motion from one video model, gain a voice from the audio stack, gain a synchronized mouth from the lip sync engine, and gain captions in the vertical video tool, without a single file ever being downloaded and re-uploaded. That is the sequence FOTOhub built the platform to make ordinary, and it is now genuinely ordinary.

Chat Live: forty-eight conversational models, and a chat that can build things

The workspace at fotohub.app/dashboard/chat-live is where FOTOhub's orchestration thesis is most visible to a non-technical user, because it puts fourteen model families in one dropdown and asks the user to think about nothing else. Google Gemini, Anthropic Claude, OpenAI, xAI Grok, DeepSeek, Alibaba Qwen, Mistral, Amazon Nova, MiniMax, Moonshot Kimi, Z.AI, NVIDIA, open-weight releases, and FOTOhub's own FH Gabriel model running on the company's own GPUs are all reachable from the same conversation. Switching between them mid-thread does not reset history, lose attachments, or change how the interface behaves. Reasoning-capable models expose their thinking where the provider permits it. Vision-capable models accept images, documents, and PDFs. Every model is labeled with the plan tier that includes it, and the free tier includes a genuinely free conversational model, because FH Gabriel runs on hardware FOTOhub owns and therefore carries no per-token invoice to pass along.

What makes this more than a model switcher is that the chat is wired into the rest of the platform. It searches the web when a question needs current information, runs deeper multi-step research when a question needs more than a search, remembers durable facts about a user across conversations, keeps a per-user file workspace so generated assets accumulate somewhere findable, and generates images inline from the same registry the generator uses, with the same pricing and the same quality. It also connects to external tool servers, which is the quiet feature that turns a chat window into an operations console: a business can attach its own systems and let the conversation act on them, with FOTOhub handling the transport, the authentication, and the accounting.

Streaming was rebuilt for this release, and the reason is worth stating plainly, because it is the difference between a chat that feels alive and a chat that feels broken. A response that arrives as one block after nine seconds and a response that begins arriving after four hundred milliseconds contain identical text and produce entirely different products. FOTOhub now streams from first token across every provider that supports it, including through the public API, so applications built on the platform inherit the same responsiveness the dashboard has.

Canvas board: the workspace where generated assets become a design

Generation produces assets. Design produces outcomes, and the gap between the two is where most AI creative tools stop. FOTOhub's canvas board closes it. It is an infinite, multi-artboard workspace with real layers, precise object properties, text with typography controls, shapes, drawing tools, an icon library, perspective and mockup tooling, grid and alignment overlays, and export presets, and it treats every AI operation on the platform as a native canvas action rather than an import step.

An object on the canvas can be generated in place, replaced by a variation, stripped of its background, extended beyond its original frame, or corrected in a region, and the result lands as a layer with the same handles and the same z-order behavior as anything else on the board. Assets from the generator, the video workspace, the chat conversation, and the user's file library are all browsable from inside the canvas. An agent mode can operate on the board directly, taking an instruction about the composition rather than about a single image. Boards are projects with history, so an evolving campaign lives in one place instead of in a download folder sorted by timestamp.

This is the surface that makes FOTOhub usable as a finishing tool rather than only a sourcing tool. A social post, a product page hero, a thumbnail set, or a full brand sheet can be composed, corrected, and exported without another application in the chain, and because every AI operation available on the canvas is the same operation available everywhere else on the platform, the quality does not drop when the work moves from generating to arranging.

Flows: the same models, running without a person in the loop

Flows is FOTOhub's automation layer, and it exists because professional creative work is repetitive in a very specific way. The prompt changes; the pipeline does not. A product photography pipeline is always the same seven decisions applied to a different product. A social pack is always the same nine crops of a different asset. A real estate listing is always the same treatment of a different set of rooms.

Flows turns that shape into a reusable object. A flow is a sequence of generation and processing steps, each bound to a specific model with specific settings, that runs end to end against new inputs and produces a finished set of outputs. The platform ships a template gallery covering the patterns that come up most often in commercial work, including product photography, multi-angle scenes, fashion campaigns, e-commerce grids, social media packs, real estate showcases, food and recipe sets, brand identity kits, interior design, travel campaigns, portrait studio work, wedding highlights, automotive showcases, and album art. Templates are starting points rather than fixed pipelines, and a flow can be built from scratch, edited step by step, previewed before it runs, and monitored while it executes, with per-step results visible as they complete rather than only at the end.

The reason flows belong in the same release as the stability work is that automation is where a fragile platform becomes an expensive platform. A pipeline that runs once with a human watching can survive an occasional silent failure. A pipeline that runs two hundred times overnight cannot. Flows run on the same reserved-then-settled accounting and the same durable job tracking as everything else, which means a failed step refunds rather than vanishes, and a completed run can be audited afterward against what was actually charged.

FH Shorts: a vertical video factory, not a template filler

FH Shorts is the newest of the major tools and the most opinionated. It takes long-form source material, an article, a recording, a podcast episode, a long video, and produces publication-ready vertical clips with captions, supporting footage, and platform-appropriate framing.

The pipeline is genuinely a pipeline rather than a single call. Source material is analyzed for the moments worth extracting. Clips are cut and framed for vertical delivery. Captions are generated with word-level timing and styled through a preset system that supports karaoke-style highlighting, with an editor for correcting both the words and the timing, because automatic transcription is very good and not perfect, and a caption with the wrong word in it is worse than no caption at all. Supporting B-roll is suggested and managed per clip. A series splitter turns one long source into a sequenced set of posts. Batch rendering produces the whole set at once. Analytics close the loop by feeding performance back into which clips get made next, and a publishing step delivers the finished output rather than leaving it in a downloads folder.

What makes this a product rather than a feature is that it is the first FOTOhub tool that is explicitly about distribution rather than creation. Everything else on the platform helps somebody make an asset. FH Shorts helps somebody run a channel, which is a different business problem and a considerably more valuable one to solve, because a channel is a recurring obligation and a recurring obligation is what makes software indispensable rather than occasionally useful.

Why the orchestration layer is the actual product

It would be reasonable to read the preceding sections as a list of six tools and conclude that FOTOhub is a suite. It is more accurate to say FOTOhub is one system with six entrances. Every surface described above reads from the same model registry, prices through the same accounting layer, writes into the same asset library, and returns through the same delivery contract. Add a model once and it appears everywhere. Fix a provider quirk once and every surface benefits. Change a price in one place and every entrance charges correctly, including the public API.

This is why the model count is worth stating and why it is not the point on its own. Two hundred and twenty-six models reachable through six unconnected tools would be a catalogue. Two hundred and twenty-six models reachable through one orchestration layer is a platform, because it means the models compose. An image from one vendor can be animated by a second, voiced by a third, captioned by a fourth, and arranged by a fifth, and at no point does the user handle a file, a credential, or a format conversion. The composition is the value that no individual model vendor can offer, and it is structurally unavailable to anyone who has integrated fewer providers or integrated them less carefully.

It is also why the platform can move as fast as it does on new releases. When a major new image or video model ships, the work of adding it to FOTOhub is registry work rather than product work, which is how newly released models tend to appear on the platform within days of becoming available, already priced, already health-monitored, already usable from the generator, the chat, the flows, and the API on the same afternoon.

Three segments, deliberately separated

FOTOhub now develops along three tracks that share infrastructure and share almost nothing else, because the mistake most creative AI companies make is trying to serve all three with one interface.

The consumer track is the dashboard and the studio applications. Its measure of success is whether somebody with no technical background and no creative training gets a result they are proud of on the first attempt. That is what the hundred-plus purpose-built applications exist for, and it is why the free tier includes a real conversational model and real generation credits rather than a countdown timer.

The business track is teams, shared budgets, role-based access, a shared asset library, brand consistency tooling, and the volume pricing that makes a creative department's monthly output predictable rather than alarming. Its measure of success is whether a marketing team can replace a scattered stack of subscriptions with one line item and lose nothing in the process. Server-enforced shared credit pools mean a team's spend is governed centrally rather than trusted individually, which is the requirement that turns a tool into something a finance department will approve.

The developer track is the public REST API, the official Python and TypeScript client libraries, signed webhooks, tool-server integration, and the e-commerce bridge, all documented at docs.fotohub.app across a documentation site that now runs to dozens of reference pages. Its measure of success is whether an engineer can go from signup to a working generation in a handful of minutes without talking to anybody. The API exposes the full catalogue, not a reduced subset, which is the difference between an API that is a marketing artifact and an API that is a product. The e-commerce bridge extends that to storefronts directly, so a shop can generate and refresh product imagery as part of its own catalogue workflow rather than as a separate manual chore, through integrations that sit natively inside the commerce platforms merchants already run.

Amazon Web Services, and what sixteen regions actually change

FOTOhub is now part of the Amazon Web Services partner portfolio, and the integration runs considerably deeper than a logo on a slide. The developer console at fotohub.app/console provisions real AWS infrastructure on a customer's behalf, in sixteen regions across North America, South America, Europe, and Asia Pacific, with the credentials for that infrastructure held and rotated by the platform so the customer never handles them.

Storage is real object storage with real controls: lifecycle policies, versioning, encryption including customer-supplied keys, object lock in both reversible and immutable modes, access policies, and event notifications that let a bucket feed whatever pipeline a customer already runs rather than only the one FOTOhub built. Output routing connects that storage back to generation, so a video model's results can land automatically in one bucket while an image model's results land in another, in a customer's own region, under a path layout the customer defines. For teams with data residency obligations, that is not a convenience feature. It is the precondition for using the platform at all, and it is why regional coverage is being reported as a product capability rather than an infrastructure footnote.

Cloud computing and GPU capacity sit alongside storage in the same console, for customers whose requirements go past calling an endpoint. FOTOhub already operates its own GPU fleet for the models it runs in-house, including FH Gabriel and the image and audio tooling that benefits from dedicated hardware, and the console exposes rentable instances, a hardware catalogue, and monitoring for customers who need the same thing under their own control. The combination, a managed model catalogue on one side and raw capacity on the other, is unusual for a creative platform and is the direct consequence of building on AWS rather than reselling somebody else's abstraction over it.

926,000 users, 46 countries, and what comes next

The platform now serves 926,000 registered users across 46 countries. Approaching a million people is a milestone worth naming, but the number that shapes the roadmap is the second one, because 46 countries means the platform is already operating as international infrastructure rather than a regional product, with the pricing, currency, language, latency, and data residency consequences that follow from that.

The direction from here follows the same logic as this release. More models, added through the registry rather than bolted on. Deeper composition between the tools, so that an asset moving from generation to arrangement to distribution crosses fewer seams. More automation, so that the pipelines professionals build once run reliably a thousand times. More regional coverage, so that a customer's data sits where their regulator expects it to sit. And continued expansion of the developer platform and the e-commerce bridge, because the largest volumes on any creative AI platform will ultimately arrive through an API rather than a browser, and the platform that handles that volume correctly, with honest pricing, durable jobs, and refunds when a provider fails, is the platform that keeps it.

FOTOhub is available at fotohub.app, with the free tier requiring no payment details. Developer documentation, API reference, client libraries, and e-commerce integration guides are at docs.fotohub.app.

About FOTOhub: FOTOhub is a creative AI platform giving individuals, businesses, and developers access to 226 AI models for image, video, audio, conversation, text, virtual try-on, and 3D generation through one orchestration layer, one balance, and one API. Founded in Poland and built on Amazon Web Services, FOTOhub serves 926,000 users across 46 countries. Media enquiries: [email protected]