Sep 25, 2026

OpenCreator: A Local-First AI Creator Workspace with 7 Video Skills

Run an open-source creator workspace on Codex CLI — video translation, dubbing, and thumbnails driven by reusable agent skills, with data staying local by default.

#tutorial#developer-tools#creative#ai-agents

Creator work sprawls across tools: one app to translate a video, another to dub it, a third for thumbnails, and none of them talk to your agent. OpenCreator — the 12k-star open-source workspace formerly known as KrillinAI — puts visual creator tools, an agent conversation, and reusable Skills in one local-first app built on Codex CLI.

Why This Skill Matters

Most AI creator apps are hosted services: your footage and scripts leave your machine, and the workflow is whatever the vendor's UI allows. OpenCreator takes the opposite design. It reuses the Codex Agent loop as the execution engine, adds a local Runtime that keeps projects, runs, and approvals on your machine, and wraps both in a visual workspace and an Electron desktop host. The daemon listens only on 127.0.0.1 and every API except the health check requires a bearer token, so the local boundary is explicit rather than implied.

For skills specifically, OpenCreator ships 7 video-production skills in the repository and reads your existing Codex skills from $CODEX_HOME — so the skills you already maintain keep working inside it.

Installation

The desktop app is the zero-setup route. Download the installer for your platform from the latest release — macOS Apple Silicon, macOS Intel, or Windows x64 — and open it. No Node.js or pnpm needed: the package bundles Codex CLI, starts the local runtime on first launch, and prepares a default project. Real model tasks require a valid Codex login.

For development or browser use, you need Node.js 22 or later, pnpm 9.15.0, and a Codex CLI executable with a valid login:

git clone https://github.com/krillinai/OpenCreator.git
cd OpenCreator
corepack enable
pnpm install
pnpm web:dev

Open http://127.0.0.1:19861/, then confirm an existing ChatGPT sign-in or API key, or point it at an OpenAI-compatible provider.

Real Workflow: Translate a Video into Bilingual Subtitles

Video Translation is the most documented workflow in the README. Import a local or public video, then let the workspace transcribe it with a cloud or local Whisper service. From there, the LLM context handles subtitle segmentation, alignment, terminology, and translation, so you can configure bilingual subtitles, add dubbing from a custom voice sample, set subtitle styles and landscape or portrait composition, and export SRT, audio, or final video.

The skills side of the same pipeline lives in the repository's skills/ directory. Seven skills cover the stages:

Skill	What it does
KrillinAI CLI	Chooses commands and interprets progress, manifests, outputs, and errors
Subtitle	Downloads platform captions or transcribes media, then translates into bilingual or short portrait subtitles
TTS	Generates target-language dubbing, optionally producing a dubbed video
Landscape Render	Renders landscape videos with bilingual subtitles or dubbed audio
Portrait Render	Composes portrait videos with titles, bilingual subtitles, or dubbing
Cover	Generates a cover image from a complete text prompt
Pipeline Plan	Validates a multi-stage output plan in dry-run mode; actual work runs through the individual stage skills

In the agent conversation, select a skill from the composer with / so the next task follows its workflow, and browse the skill marketplace and installation history in the plugin center.

The Rest of the Workspace

The README counts ten creator tools in the current release — video translation and downloading, thumbnail generation, image generation, article writing, Xiaohongshu posts, short-video scripts, stick figure animation, smart dubbing, and video generation — with two more (Auto Clips and Digital Avatar) marked in development. Every revision creates a new version instead of overwriting, and creation templates give you a starting point without building prompts from scratch.

Tips

Image, video, voice, and transcription models are configured in Settings → AI Services; some providers, like Edge TTS, need no API key.
OpenCreator checks the bundled yt-dlp for updates every seven days but never installs them automatically — updates are an explicit user action, and the working version stays available if one fails.
Set CODEX_HOME if you want an isolated Codex environment; OpenCreator uses the active one by default, so confirm the impact before changing global skills or MCP configuration.
Sensitive memory requires a second confirmation before it is stored.

When Not to Use This

Real model tasks require a valid Codex login, and video workflow skills require the CLI and relevant services to be configured — inclusion in the repository does not mean every service is bundled. If your creative work is image-only and already lives happily inside a hosted tool, a local workspace with a Codex dependency is extra machinery. And if you need auto-clipping today, it is an in-development item, not a shipped feature.

See the leaderboard for more skills.