GetStream/Vision-Agents
Open Vision Agents by Stream is a platform for building real-time, low-latency multi-modal AI agents that understand and interact with video content using advanced models and Stream's edge network.
Awesome AI Agents › Video Processing Agents
nexu-io/html-video
html-video is an Apache-2.0 licensed project from the Open Design team that lets a local coding agent turn a description, an article link or a GitHub repository into an animated, multi-frame HTML video and render it to a real MP4 file on the local machine. The workflow runs as a single agent loop: the agent selects a rendering engine, picks a template from a curated gallery of 21 animated single-file HTML templates, fills the template with the supplied content and exports the result. Rendering is handled by headless Chromium, which records the animated HTML frame by frame, and by ffmpeg, which encodes the frames with libx264. The project is designed as a meta layer above existing HTML-to-video engines: each backend implements one render(input, ctx) adapter contract, so adding an engine makes it available to every template, every agent and the whole studio workflow without introducing a new authoring DSL. According to the README the Hyperframes engine, based on HTML, CSS and GSAP, is the only fully wired backend today, while Remotion, Motion Canvas, Revideo and Manim adapters are planned or under research. Fourteen coding agent backends are supported and auto-detected on PATH, including Open Design, Windsurf CLI, Trae CLI, Claude Code, Cursor Agent, Codex CLI, Gemini CLI, Grok Build, Qwen Code, OpenCode, GitHub Copilot CLI, Aider, Hermes and the Anthropic Messages API, and the active one can be switched from the top bar. Templates cover editorial data charts, glitch and kinetic title cards, liquid gradient heroes, cinematic light-leak frames, typewriter VFX, logo outros, multi-scene product promos and decision-tree explainers. An optional AI soundtrack adds music and narration. Everything runs locally with no per-render fees and no vendor lock-in, which suits developers and content teams who want programmatic video generation driven by an agent rather than a hosted service.
https://github.com/nexu-io/html-video
Also in Video Processing Agents
RayVentura/ShortGPT
ShortGPT is an AI-powered framework that automates video content creation and editing for platforms like YouTube Shorts and TikTok, supporting multi-language voiceovers and automated footage sourcing.
showlab/Awesome-Video-Diffusion
A curated list of recent diffusion models and resources for video generation, editing, restoration, and various AI-driven video applications.
dramaclaw/dramaclaw
Source-available AIGC video production pipeline that turns a manuscript into a finished short drama, handling character extraction, episode planning, scripts, storyboards, voice-over and final editing.
HKUDS/VideoAgent
Multi-agent framework for video understanding, editing and generation that parses user intent, builds a graph-based tool workflow and produces edited or newly created videos from natural language.
video-db/Director
Director is an AI video agents framework that enables complex video tasks like search, editing, and generation through natural language commands and real-time streaming, built on VideoDB's video-as-data infrastructure.
video-db/StreamRAG
StreamRAG is a GPT-powered video search and streaming agent that enables real-time video search, summarization, and publishing of searchable video collections on the ChatGPT store.
microsoft/DeepVideoDiscovery
Microsoft research agent for long-form video question answering that treats segmented clips as an environment and iteratively selects multi-granular search tools, reaching state of the art on LVBench.