← all challenges

// challenge brief · hack a ton 2026

Savoy Content Studio

A no-code studio that lets a hotel's marketing team turn a text idea into a finished 9:16 reel, starring their locked cartoon mascot, without touching code.

// Proposed by
Savoy Hotel Mamaia
// Industry
Hospitality / content & social media
// Difficulty
🟡 Intermediate

★ Cash prize plus a paid collaboration to build the full V1 after the hackathon.

Read the sponsor's original brief, with the full spec and resources.Download →

// the problem

Savoy Hotel Mamaia (143 rooms, all-inclusive) has a brand character: "Domnul Girafă", a cartoon giraffe that already lives on Instagram, TikTok, and Facebook. The plan is ambitious: 30+ scenes and 4 reels a week, with a strict visual identity and a voice already configured in ElevenLabs.

The catch is the workflow. Producing this content today means Claude Code, Python, and a stack of APIs. The marketing team does not have those skills, and shouldn't need them. There is no simple tool the team can actually use day to day, so the content pipeline stays stuck on whoever can run scripts.

// your mission

Build a web platform with a login that a non-technical marketer opens, types an idea, approves a preview, and downloads a finished short video. Two modes cover the hotel's content. The hard constraint runs through everything: the giraffe is a locked visual asset and must never be redesigned by the AI.

// how it works

The platform has two modes.

  1. 1Mode 1 — Hotel scene (no giraffe): the user writes a free prompt or picks one from a library. The system generates an image, the user approves it, then it produces an ~8s video from that image.
  2. 2Mode 2 — The giraffe speaks: the user writes the dialogue line plus a scene description. The system generates an image of the giraffe in that scene with its identity locked, the user approves, then it generates the ElevenLabs voice and a lip-synced video.
  3. 3Approve before you spend: image approval gates the expensive video step, so the team never burns generation budget on a take they don't want.
  4. 4Final format: 9:16, 15–30 seconds, ready to download and post.

// what we provide

Everything needed to build against real assets during the hackathon.

  • The mascot master image (girafa.jpg, 686×1099) as the single source of truth for the character.
  • The locked visual identity, to be pasted verbatim into every prompt alongside "DO NOT redesign — LOCKED visual asset":
  • Body: glossy neon-yellow / gold.
  • Spots: orange, on the legs, back, and neck.
  • Hat: beige/brown straw hat, tilted slightly for comic effect.
  • Bow tie: orange to red-orange, at the neck.
  • Eyes: large and expressive, blue pupil, white sclera, sweet look.
  • Outline: clean black, 90s–2000s cartoon style.
  • Proportions: very long neck (~40% of height), small body, black hooves.
  • The voice, already configured in ElevenLabs: Voice ID g8YRbOlJsPkrezcSUiCM, model eleven_multilingual_v2, stability 0.5, similarity boost 0.75, style 0.0, speaker boost on. Tone: mature, warm, pragmatic and jovial. Language: Romanian.
  • Active API keys for the duration of the hackathon: Higgsfield, ElevenLabs, HeyGen.
  • A `config.yaml` with the existing project parameters.
  • Hetzner infrastructure available for deploy and testing.

// suggested approach

  • Recommended stack: FastAPI (Python) backend, plain HTML + Bootstrap frontend (no React), deployable on a Linux server (Hetzner).
  • Identity is the whole game. Image models will quietly redesign the character unless told not to. Always send the locked description plus the "DO NOT redesign" instruction, and prefer an identity-preserving image step (e.g. Nano Banana) for the giraffe.
  • Pipelines per mode: scene image (Higgsfield / GPT Image) → approve → video (Seedance) for Mode 1; locked giraffe image → approve → ElevenLabs voice → lip-sync video (HeyGen / Kling) for Mode 2.
  • Floor version: login, the giraffe mode (dialogue text → identity-locked image + ElevenLabs voice), the hotel mode (prompt → Higgsfield image), and a download button. Add lip-sync, the prompt library, and the approval gate from there.

// how you'll be judged

  • Could a non-technical marketer actually use it end to end, with no code?
  • Identity fidelity: does the giraffe stay on-model every time?
  • Voice quality and correct Romanian delivery.
  • Does the full pipeline work live, from text to a downloadable 9:16 video?
  • Cost discipline: does approval gate the expensive steps?

// stretch goals

  • Lip-sync video so the giraffe actually moves its mouth (HeyGen / Kling).
  • A library of pre-built prompts the team can reuse.
  • Image approval before video generation as a hard cost guard.

// deliverables

  • Git repo: code + README + run instructions.
  • Demo: a finished reel produced live on stage, in both modes.
  • 5-minute pitch: architecture, decisions, trade-offs.