CineCaption AI Multi-Style Video Captioning Agent

Created by team OnePiece on July 07, 2026
DeepSeek V3
Video Captioning

CineCaption AI is a video captioning agent built for the AMD Developer Hackathon Track 2. It downloads a video clip, extracts eight evenly-spaced key frames using ffmpeg, and sends them to Kimi K2.6 — a vision-language model on Fireworks AI — which produces a detailed factual description of the scene. That description is then passed to DeepSeek V4 Pro, a text model that generates four styled captions: formal, sarcastic, humorous-tech, and humorous-non-tech. The pipeline processes multiple videos in parallel using ThreadPoolExecutor, completing each clip in 30-60 seconds. It includes automatic retry logic, JSON parsing with fallback, and a per-style fallback caption generator to guarantee output for every requested style. The entire agent runs inside a lightweight Docker container (python:3.11-slim + ffmpeg) and is deployed to GitHub Container Registry with a linux/amd64 manifest.

Category tags:
GithubPresentationDemo
Explore more applications

Thymus

Thymus is a lightweight hybrid token-efficient router designed to maximize accuracy while minimizing token costs in multi‑task LLM pipelines. It dynamically routes user queries across local and remote models on LLM providers.

The Disappointer

HuggingFace HubLLaMAAMD Developer Cloud
Vercel
application badge

AI Classroom Edge Intelligence

A privacy-first classroom AI platform that routes sensitive work to local edge systems and eligible anonymized analysis to Fireworks AI, helping teachers make faster, safer instructional decisions even with unreliable internet.

AI Classroom Edge

AMD Developer CloudQwen3rest apiGithub CopilotCodexChatGPT

Taskly: Smart Multi-Model Task Router

Taskly classifies incoming tasks into 8 categories (QA, math, code, NLP) and routes each to the optimal Fireworks AI model with a tuned prompt — maximizing accuracy while minimizing token usage in a fully Dockerized pipeline.

RuntimeTerror

AMD Developer CloudAMD ROCm
Streamlit
application badge

router_007_v3

router_007_v2 is a Track 1 agent that records **zero billable tokens**: every answer is computed inside the container by a Qwen2.5-7B-Instruct model bundled in the Docker image and served in-process with llama-cpp-python

roc_auc_half

Claude CodeChatGPTAMD ROCmAMD Developer CloudCodexGemmaGPT-5NVIDIAQwen3

Lexyprep

Agent knows how to make thorough legal research based on my experience of winning legal cases. It understands your issue and advises on procedure

Lexyprep - Do you have a case

AMD Developer CloudAMD ROCmCodexGemma