Skip to content

What is AI Sticker Pack

Everyone. Read this page first, whatever your role. It needs no technical background.

  • AI Sticker Pack is an event activation: a guest takes one photo at a tablet station and receives a personal Telegram sticker pack — 12 cartoon stickers with their own face — a minute or two later.
  • The guest needs only Telegram. There is no app to install and nobody waits at the station.
  • The organiser gets a branded station that runs itself: consent is collected in the bot, photos are deleted automatically within 72 hours, and each Telegram account gets exactly one pack.
  • A separate landing page collects organiser requests (“leads”) and posts them as cards into a Telegram group where the sales team works them.
  • The product has run as a pilot for a single station of roughly 70 guests. Everything on this site describes the code as it is, including what is not finished — see Known gaps.
flowchart LR
  subgraph Venue
    G[Guest with a phone]
    K[Tablet kiosk]
  end
  subgraph Server["Our server"]
    API[Kiosk API]
    BOT[Telegram bot and generation worker]
  end
  TG[Telegram]
  AI[AI image provider]
  G -- "scans QR, agrees" --> TG
  G -- "taps Start, picks outfit, poses" --> K
  K -- "session and photo" --> API
  API -- "job in queue" --> BOT
  BOT -- "draw sticker sheets" --> AI
  BOT -- "create sticker pack, send link" --> TG
  TG -- "pack link" --> G
  • The event stays in conversations. Photos sit in a camera roll; stickers get sent in chats long after the venue is empty. This is the main selling argument on the landing page.
  • Nothing to run. The station is delivered, set up, rehearsed and monitored for them. They do not collect consent on paper, store photos or manage a queue.
  • Their look. The station screen carries the event’s colours and logos. Outfits on the stickers come from the product catalogue.
  • Predictable privacy story. Consent first, automatic deletion, only service data kept.
  • A personal pack of 12 stickers in a “chunky cartoon” style that carries their features: hairstyle, skin tone, facial hair, face shape.
  • About one minute at the station: tap, scan, agree, pick an outfit, pose.
  • The pack arrives in the Telegram chat with the bot and is added with one tap.
# Where What happens
1 Tablet The idle screen shows a Start button. The guest taps it; only now a session and a QR code are created.
2 Phone The guest scans the QR code with the phone camera. Telegram opens the bot with the session token.
3 Bot The bot checks whether this Telegram account already has a pack. If not, it shows the consent text. The guest taps Agree.
4 Tablet Only after consent the tablet shows six outfit cards and warms up the camera. The guest picks an outfit.
5 Tablet The tablet counts down, shoots a burst, uploads the best frame and immediately returns to the idle screen for the next guest.
6 Server A background worker draws the stickers with an AI image model, cuts and finishes them in code, and creates a Telegram sticker pack.
7 Bot The bot sends the link only after the pack is verified to be really available. Measured time from photo to link in the server load test: median 56 seconds, about 1–1.3 minutes with real Telegram delays (load test report).

Step-by-step with code references: Guest session flow, Generation flow, Pack delivery flow.

# Where What happens
1 Landing page An organiser fills a short form: name, Telegram, phone, optional date, guest count and comment.
2 Landing server The form is validated, spam checks run, and the request is sent to the leads endpoint of the product server with a shared secret.
3 Product server The request is stored in the database and answered at once. Telegram is not involved yet.
4 Bot A background loop publishes the request as a card in the sales Telegram group, with status buttons.
5 Sales group Lead admins press In work, Contacted, Deal, Refused or Spam. The bot reminds about untouched cards and posts a morning digest.
— Fallback If the product server does not answer, the landing page posts the request straight into the group with a warning line. Nothing is lost, but that request is not in the database.

Details: Lead submission flow and How a request is handled.

These are hard rules in the code base, not preferences. Changing one is a product decision, not a refactoring.

Principle What it means in practice
Consent before the photo The tablet cannot show outfits or take a photo until the guest has tapped Agree in the bot. The API answers 409 otherwise.
Photos live at most 72 hours Selfies and every file derived from them are deleted automatically. The configured lifetime must be between 1 and 72 hours or the service refuses to start. The database never stores an image, only metadata.
One pack per Telegram account Checked when the guest opens the bot, when they pick an outfit, and atomically when the job is queued. An admin can grant one more attempt.
The station never waits for generation The photo upload returns immediately; delivery is asynchronous through the bot.
Two AI providers by design The pipeline takes the provider from configuration and can fall back to a second one.
Finishing is code, not AI Cutting figures out of the sheet, the white sticker outline and export to Telegram’s format are deterministic code.

Why each rule exists: Decisions (ADR).

  • No pack without Telegram. A sticker pack lives in Telegram; there is no WhatsApp, e-mail or print delivery.
  • No offline mode. The tablet, the server, Telegram and the AI provider all need the internet.
  • No guarantee of exact likeness. The result is a stylisation that carries the guest’s features, not a photo.
  • No branded outfits or props. Only the station screen is branded; outfits come from a fixed catalogue.
  • No choice of art style by the guest. One style is configured per installation.
  • No second pack on request. Only an admin can allow another attempt.
  • Not a CRM. The leads module stores requests, statuses and basic statistics. It has no pipelines, e-mail or integrations.
  • Not multi-tenant yet. One installation serves one bot, one event configuration and one sales group. See Multi-tenant and white label.
  • Bot and kiosk speak Russian only. The landing page has Russian, Uzbek and English.
  • Paths by role — what to read next for your job.
  • Glossary — kiosk, session, job, sheet, anchor and the rest.
  • System context — the same picture with every external system.

Last verified: 2026-11-13 against ai-sticker-pack@b715418 and ai-stickers-landing@e350ff5.