Consumer generative-content studio inside a Telegram Mini App

Context

Consumer AI content is scattered across a dozen single-purpose tools: one site for images, another for music, a third for avatars — each with its own login, its own credits, its own learning curve. For a non-technical user the friction is not the model quality, it is the assembly: finding the right tool, paying it with a card it will accept, and stitching results together.

The opportunity was to collapse that stack into one place people already live in — a Telegram Mini App — where a trend catalog does the deciding, a single credit balance does the paying, and every generator hides behind the same three-tap flow.

Approach

The core product decision was to build a consumer product, not a model playground. The Mini App opens on a catalog of live trends and templates; picking one drops the user straight into a generation screen with sensible defaults. Discovery is carried by the catalog rather than a blank prompt box: the user picks an outcome they have already seen work, and the system supplies the model, the parameters and the price.

That framing dictated the engineering priorities: the volatile part — which model is best this month — had to be swappable behind a stable interface, while the durable parts — catalog, billing, cabinet, admin — had to evolve on their own release cadence.

How it works

Behind the screens, a routing layer dispatches each job to the appropriate provider — image models via kie.ai and the frontier APIs, Suno for music, Kling and Veo for video, OpenAI, Anthropic and Gemini for text — and normalizes seven very different APIs into one job model with a shared credit cost and a single generation history.

The backend is FastAPI over Supabase; the frontend is a Telegram Mini App, so identity, distribution and payments all ride on a platform the user already has. Nothing is installed, and the account is the Telegram account.

Key engineering details

Eight product modules surround the generators: the trend catalog, onboarding, generation screens, tariffs, a personal cabinet with history, the credit economy, an admin panel and a content system that lets operators grow the catalog without shipping code. Billing is deliberately dual-rail — local bank cards and Telegram Stars — but both settle into the same credit ledger, so every generator prices in one unit regardless of how the balance was funded.

Provider volatility is treated as a given, not an exception: models are added, swapped or retired behind the router without touching the product surface, and each job's credit price is set independently of the provider's own billing granularity.

Outcome & what shipped

The platform runs in production with its modules live and expanding: six content types, seven routed providers, a growing trend catalog and a unified credit economy behind one Mini App. It also serves as the host platform for the studio's adjacent deliverables — the marketing site, the custom-song ordering module and the video-tutorial factory each plug into surfaces this platform defines.

As a case, it demonstrates the studio's core competence: wrapping a volatile multi-provider generative backend in a stable, monetized consumer interface — and keeping catalog, billing and admin evolving independently of the models underneath.

What we built

• Trend catalog

  A curated, data-driven catalog of trends and templates as the entry point — updated through the content system, no deploy required.

• Credit economy & billing

  One credit balance spans every generator; each job carries a shared, transparent cost regardless of which provider fulfills it. Two payment rails — local cards and Telegram Stars — fund the same ledger.

• Multi-provider generation router

  A single job model dispatches to seven image, music, video and text providers and normalizes their responses back into one history.

• Admin & content system

  Catalog, tariffs and content managed from an admin panel so operators expand the product without engineering.