Media — generate images, videos, and audio right inside the environment
Image, video, and audio without leaving the cockpit — 4 native media MCP tools.
The studio inside the environment
Overclock has 4 native media MCP tools: generate image, generate video, generate audio, and list the enabled providers. They're part of the embedded server's 52 tools — meaning any authorized agent generates media from inside the cockpit, mid-work, without exporting anything to another tool.
The media providers supported today, each with its own key: Gemini and GPT Image for images, Replicate for video (the Kling, Seedance, Wan, and Minimax families, with keyframe interpolation), and ElevenLabs for voice and audio. <!-- âncora: verificado no código do app 2026-07-10 — packages/core/src/ServiceKeysStore.ts (google-image, openai-image, replicate-video, elevenlabs) + mediaGeneration.ts -->
It's the base for the creative squads: the cinema team breaks a brief down into shots, generates each scene through keyframes, and assembles the score — all through these 4 tools, with the formation's allowlist guaranteeing no one swaps engines midway.
How to use it
Configure the providers
image, video, and audio service keys in Settings, encrypted at rest.
Run a creative squad
the formation already declares which providers and models to use; the agents generate inside the mission.
Get the files in the workspace
image, clip, and score land in the project like any other artifact from the team.
What this area has today
Image generationshipped
Real text-to-image and img2img — editing that preserves the reference face, with a real aspect ratio.
Video generationshipped
Clips with keyframe interpolation — start frame + end frame, the model animates the middle.
Audio generationshipped
Voice and effects from a prompt — the cinema squad's score comes from here.
Allowlist per formationshipped
Every squad declares its media providers — no agent swaps engines silently.
Multi-model per providershipped
Several models enabled per provider, with one marked as default — generation respects the list.
Encrypted keysshipped
Media keys stay encrypted at rest (AES-256-GCM) — out of reach of the panes.
Status straight from the product's brain: shipped = actually in use; partial = exists, with documented limits.
Where this actually gets to work
all squads →Arcade
Jogável em uma sessão. Com vídeo do bot jogando pra provar: mecânica aprovada no papel, código e assets em paralelo, playtest em 3 lentes — 7 panes, 3 gates.
Overclock OriginalCinema Site
Rouba o DNA da referência. Entrega uma produção de cinema: site completo com design extraído ao vivo, copy por seção e 6 imagens geradas em paralelo — pico de 9 panes, 2 gates.