waoowaoo: A Self-Hosted AI Creative Workspace for Images and Videos

What waoowaoo Does and Who It Is For

waoowaoo is an AI creative workspace where users develop image and video concepts with a right-hand assistant, upload reference material, and organize generated results on a canvas. The README describes it as an ongoing project context: the assistant helps develop a brief, and assets are generated within that project.

Generation requests go through an OpenRouter API key supplied by the user. The README states that provider calls are paid through the user's own account. Available models in the preview include GPT Image 2, Nano Banana variants, Seedance variants, and MiniMax H3 and H3 Max. The application interface supports Chinese and English; the Japanese and Korean README files are translations, not full UI localizations.

The project is described in the repository description as an industry-first professional AI agent platform for controllable film and video production, supporting workflows from shorts to live-action at Hollywood-standard quality. The README is more measured: it describes the current preview as an AI creative workspace iterating rapidly with some bugs remaining.

Installing waoowaoo with Docker Compose

The recommended installation path uses prebuilt Docker images via Docker Compose. The README describes this as the recommended distribution; you do not need to install Node.js, MySQL, Redis, Temporal, or FFmpeg on the host machine.

The README's Quick Start section provides a text prompt to copy into a coding assistant (such as Claude Code, Cursor, or Codex CLI) that automates the setup:

Install the self-hosted preview of https://github.com/waooAI/waoowaoo on my computer.
Read the repository README, docs/INSTALL.md, and the selected GitHub Release first.
Use a published preview release and its prebuilt Docker images. Pin both the
application and the Codex runtime to the release's immutable image digests.

The docker-compose.yml and docker-compose.self-hosted.yml files together make up the self-hosted Compose configuration. Environment variables are set in a .env file based on .env.example. Required secrets include DATABASE_URL, COMPOSE_DATABASE_URL, MYSQL_ROOT_PASSWORD, REDIS_PASSWORD, NEXTAUTH_SECRET, CRON_SECRET, API_ENCRYPTION_KEY, and MINIO credentials.

The Compose setup starts Caddy automatically as a reverse proxy. After approving local certificate trust, the application is available at https://localhost:1443; port 13000 only redirects HTTP to HTTPS.

For source development, the README notes that npm run dev starts a local Next.js server at http://localhost:3001.

The Canvas, Assistant, and Generation Features

The README describes five core capabilities of the current preview. First, an ongoing project context: users work with the right-hand assistant to develop a brief and create assets across sessions. Second, a canvas for organizing files, uploading reference material, and inspecting generated images and videos. Third, model-aware generation with aspect ratios, durations, and resolutions adapted to the selected model. Fourth, reference modes including first-frame, first/last-frame, and reference-image, with availability depending on the selected model. Fifth, refinement of an existing result into a new version while retaining the original.

The README notes that MiniMax H3 Max accepts a single first frame in this application rather than a first/last-frame pair in the current preview. Music and voiceover controls are not included in the preview.

Tech Stack and Local Data Storage

The README lists the technology components. The framework is Next.js 16 with React 19. The database is MySQL managed through Prisma ORM. Durable workflow execution uses Temporal. Transport and caching use Redis. Media files are stored in a private MinIO instance, using the S3 API. The front end uses Tailwind CSS v4. Authentication uses NextAuth.js.

The docker-compose.yml shows the required environment variables for each service: DATABASE_URL, REDIS_HOST, REDIS_PORT, S3_ENDPOINT, S3_BUCKET, NEXTAUTH_URL, and others. The self-hosted Compose overlay adds the MinIO service and its initialization step.

The README is explicit about the data model: project data and media use local database and private MinIO storage. Prompts and references are sent to the provider for the AI generation tasks, but generated media is stored locally. The README states that supported inline image references do not require a public media URL.

OpenRouter API Configuration and Model Availability

waoowaoo does not bundle any AI model weights. All generation runs through an OpenRouter API key that the user configures in the application's API configuration after launching. The README warns to keep API keys out of chat and support logs.

The models listed in the preview README include GPT Image 2, Nano Banana variants, Seedance variants, and MiniMax H3 and H3 Max. The README notes that availability and charges depend on the provider, and each model's supported reference modes (first-frame, first/last-frame, or reference-image) vary. Model availability in the application depends on what OpenRouter exposes; the README does not document a fixed model registry.

The .env.example file shows a PROVIDER_CREDENTIAL_MODE variable set to user-key for the self-hosted edition, confirming that users supply their own provider credentials rather than using a shared pool.

Limitations and the Elastic Licence

waoowaoo is not open-source software. Starting with v0.5.0-beta.1, the distribution is licensed under Elastic License 2.0 (ELv2). The README states that personal use, internal business use, modification, and redistribution are permitted under ELv2, but providing third parties with hosted or managed access to a substantial set of the software's features requires separate permission from the team. Earlier releases retain their original licences.

The project is explicitly in preview: the README warns of bugs and rough edges. Music and voiceover controls are absent from the preview. Some model reference modes are constrained; for example, MiniMax H3 Max accepts only a single first frame rather than a first/last-frame pair.

Production-grade hosted alternatives such as Runway, Kling, or similar commercial video generation services offer comparable generation capabilities without requiring self-hosted infrastructure, but they do not give users the same data control or the ability to run everything locally.

Maintenance and Contribution Model

The last push to the main branch was on 2026-09-21. The most recent release is v0.5.0-beta.1, published on 2026-09-05. Earlier stable releases (v0.4.0 and v0.4.1) were from April 2026, showing a gap before the beta series resumed.

The README contribution model is non-standard: external pull requests are reviewed for ideas but the core team implements changes internally rather than merging external PRs directly. Bug reports and feature requests are accepted through GitHub Issues. The project does not encourage direct code contributions from the community in the same way a fully open-source project would.

The Node.js engine requirement is 22 or newer (from the engines field in package.json), and npm 9 or newer is required. The Dockerfile uses node:22-bookworm-slim as the base image.

Editorial conclusion

waoowaoo suits individuals and teams who want a canvas-based AI image and video workspace that runs on their own infrastructure with their own API keys, keeping generated media in local storage rather than a vendor cloud. The Elastic License 2.0 permits personal use, internal business use, and modification, but prohibits hosting it as a managed service for third parties without separate permission. The current release is v0.5.0-beta.1; confirm that Docker Engine or Desktop, Docker Compose v2, and Git are available on the target machine before starting.

Frequently asked questions

Official sources

1. Issues
2. README
3. Releases
4. waooAI/waoowaoo on GitHub