krillinai/OpenCreator

OpenCreator – an open‑source AI‑powered creative workspace

What it is

• A locally‑run application (web + optional desktop client) that lets individuals or teams create and edit multimedia content (video, images, audio, subtitles, documents) using large language models and other AI services.
• It builds on the Codex CLI – an existing agent‑loop framework – and adds a stable runtime, a visual dashboard, and a desktop host.

Core ideas

Idea	How it works
Unified agent + UI	The same underlying state machine powers both a chat‑style conversation interface and a visual “creator tools” dashboard, so you can issue natural‑language commands or click‑through settings and the two stay in sync.
Local‑first	All data, logs, and generated assets stay on your machine by default. The runtime runs as a local daemon; the desktop client only adds OS‑level features (file picker, tray, notifications).
Codex‑native	Re‑uses the Codex agent loop, model selection, tool‑call handling, Skills, and MCP (model‑config‑profile) without re‑implementing them.
Versioned workflow	Every revision creates a new version, preserving earlier settings and outputs for easy comparison and rollback.
Managed runtime components	Bundles tools like yt‑dlp and checks for updates, allowing safe roll‑backs if an update fails.

Creator tools that ship today

• Video Translation – import a video, transcribe with Whisper (cloud or local), then let an LLM segment, align and translate subtitles; can also generate dubbed audio or export portrait‑mode videos.
• Video Downloader – parse YouTube, Bilibili, etc., show available formats, and download video or audio for later processing.
• Thumbnail Generator – feed a topic, video link, and optional reference image to produce multiple thumbnail candidates.
• Image Generation – use the “GPT Image” model (or any configured image service) to create images from prompts, with aspect‑ratio and count controls.
• Road‑map items – stick‑figure animation, auto‑clip generation, smart dubbing, full video generation, digital avatars (all listed as coming soon or in development).

How you interact

1. Start a project – the runtime creates a default project on first launch.
2. Choose a tool or start a conversation – either open the dashboard or type a natural‑language request.
3. Run a task – the agent calls the appropriate tool (e.g., yt‑dlp, ffmpeg, Whisper) and streams progress back to the UI.
4. Iterate – edit settings in the UI or ask the agent to refine results; each change spawns a new version.
5. Export – download the final media (video, subtitles, images, audio) directly from the workspace.

Supported models & services

• LLMs – any model available through the Codex catalog or an OpenAI‑compatible provider (GPT, DeepSeek, Qwen, Kimi, GLM, Grok, Doubao, ERNIE, Hunyuan, etc.).
• Image – GPT Image (OpenAI) by default; other services can be added via the AI‑Services settings.
• Voice & transcription – Whisper, OpenAI TTS, MiniMax, Microsoft Edge TTS, Alibaba Cloud Speech, plus any custom provider you configure.

Getting started

# clone and install
git clone https://github.com/krillinai/OpenCreator.git
cd OpenCreator
corepack enable          # ensures pnpm version from repo
pnpm install

# run the web UI (starts the local daemon automatically)
pnpm web:dev
# then open http://127.0.0.1:19861/

For a native desktop experience:

pnpm desktop:dev   # development mode with hot‑reload
pnpm desktop:dist  # build an installer for your OS

Prerequisites are Node 22+, pnpm 9+, and a Codex CLI executable logged in to an AI provider.

Architecture at a glance

• Frontend – React 18 + Vite (apps/web), shared between browser and Electron.
• Collaboration Core – a shared workflow state (steps, progress, versions) that both UI and conversation read/write.
• Local Runtime (daemon) – Node process exposing an API (SSE) that talks to the Codex CLI, manages projects, runs tools, stores data in SQLite/files, and handles updates.
• Media toolchain – FFmpeg, yt‑dlp, Whisper, and any configured AI services.

Who might use it

• Content creators who want AI‑assisted video translation, thumbnail generation, or image synthesis without sending raw assets to cloud‑only platforms.
• Small teams needing a reproducible, version‑controlled pipeline for multimedia projects.
• Developers who already use Codex agents and want a ready‑made UI/desktop wrapper.

Where to learn more

• Multilingual READMEs (English, Chinese, Japanese, Korean, etc.)
• Discord community and QQ group links in the badge bar
• Detailed runbooks for desktop packaging and operations under docs/operations/

---

Bottom line: OpenCreator is a genuine, open‑source AI workspace that couples a conversational agent (via Codex) with a visual dashboard for multimodal content creation, all running locally and version‑controlled. It is well‑documented, actively maintained, and positioned as a platform for extending AI‑driven creative workflows.