✦ AMD HACKATHON • TRACK 2

VIVID STUDIO

Powered by Gemma

Thumbnail and multilingual caption generator

CAPTIONS THUMBNAILS LANGUAGES
Create with Vivid Studio

Agent Execution Pipeline Architecture

Trace the data flow structure map from initial file ingest payloads straight down to the generation array blocks.

01 / INPUT INGEST

Accepts multi-format video files or raw streaming endpoints URLs.

PAYLOAD DEPACKING

02 / EXTRACT MATRIX

Deconstructs data timelines into distinct spatial video frames & audio matrices.

AMD ACCELERATED PROCESSING

03 / AGENT INFERENCE

Cognitive models process structural themes and synchronize dialogue vectors.

VLM CONTEXT CAPTURE

04 / QUAD OUTPUT

Generates conditional sets formatted into four divergent tonal responses.

READY MANIFEST MATRIX
THE_CAPTIONERResolving…ACCESSED VIA Fireworks.AI
THE_VISIONARYResolving…ACCESSED VIA Fireworks.AI
THE_ORCHESTRATORAdvanced features lockedENABLE GEMMA EVERYWHERE

Vivid Creation Workspace

One video. Four voices. Four vivid 16:9 directions.

READY
70% WORKSPACE

Formal Tone

English
REG_01_FORMALWAITING
✦Thumbnails powered by Gemma EverywhereEnable the mode to unlock four visual directions.
ACTIVE CAPTION
ENGLISH SOURCE

Select a preset and click Generate to reveal four fixed caption styles.

AWAITING PIPELINE
✦
Multilingual CaptionsEnglish captions are cached automatically.
✦Translation powered by Gemma EverywhereEnable the mode to unlock cached multilingual captions.
LOCAL SETUP

DIY

Run the full caption and thumbnail pipeline on your own machine in five steps. Docker Desktop must be running before you start.

Windows 10/11 Git Docker Desktop Python 3.10+
01 / CLONE REPOSITORY

Pull the source code repository from GitHub and enter the folder.

git clone https://github.com/starlite20/vividstudio.git cd vividstudio
02 / WORKSPACE INGEST SETUP

Create input/output directories and stage the sample configurations.

mkdir input, output -Force copy examples\tasks.sample.json input\tasks.json
03 / BUILD LOCAL DOCKER

Compile containerized stack dependencies inside Docker Desktop.

docker build -t video-caption-agent:local .
04 / RUN CAPTION PIPELINE

Mount local volume paths and trigger inference calculations on local host.

docker run --rm ` -v "${PWD}\input:/input" ` -v "${PWD}\output:/output" ` video-caption-agent:local
05 / VERIFY PIPELINE OUTPUT

Query generated results containing captions inside the results JSON.

Get-Content .\output\results.json

The Engineering Cluster

The core development engineers responsible for building the Vivid Studio acceleration ecosystem.

VIEW PROFILE ↗

Kabita Rajbanshi

Team Leader

Coordinates team operations, manages development timelines, defines product scope, handling presentation development and strategy.
VIEW PROFILE ↗

Samer Ibrahem

Frontend Developer

Builds and refines the client side user experience, styles responsive UI components, and connects sandbox interactions to the API broker.
VIEW PROFILE ↗

Suhail Sujaudeen

Applied AI Engineer

Architecting the Agentic AI Workflow, developing the Agentic Pipeline & prompt engineering to ensure best results with Gemma Integration.
TEAM MEMBER

Shad Obied

Lead Pipeline & QA Engineer

Oversees UI components mapping workflows, contract assertion tests, cluster infrastructure integrations, and pipeline verification protocols.