Documentation

An agent is a graph you build once and run as often as you like. Each block does one thing, and the line between two blocks carries a value: a piece of text, an image, a video. Run it and the agent walks the graph in order, producing a finished vertical video at the end.

Blocks only connect when their types match. A block that outputs an image will not plug into an input that expects text, and the canvas dims everything incompatible while you drag, so a graph that connects is a graph that runs.

• Start with gives the agent its raw material: your product, a written prompt, a reference image, a screen recording of your app, or a file you upload.
• Create with AI writes the hook and the script, generates the image, turns it into video, reads the script aloud, or makes a face speak it.
• Edit and format joins clips, burns captions, adds an overlay and locks the output to 1080x1920.
• Publish saves the video or posts it to a connected account, right away or in the next free slot.
• Advanced holds what is not available yet. A block marked so does nothing to your video, and says it on its own card.

Every block shows its own cost on its card, and the total updates as you build. You always see the price before you spend it.

A credit maps to one cent of what the provider charges. That is how a block gets its price: a video model billed at $0.112 a second costs 11.2 credits a second, with nothing added on top at the block level.

The cost of a video block depends on its model, its length and its resolution, because that is how the providers bill. A five second clip on a fast model costs around fifty credits; the same clip on a premium one can cost three times that. The estimate is computed from the exact parameters that will be sent, including the duration the model will round to.

New accounts start with 150 credits and no card. Credits reset monthly on a paid plan rather than accumulating.

Every image and video model on the platform is available to every plan. The expensive ones are not gated, they simply cost more credits, so the choice stays yours.

Models differ in ways a price list does not show: some preserve the grain of skin while others smooth it, some render hands cleanly while others do not, some refuse photorealistic faces outright. Our comparisons live on the blog.

Connect an account once in Accounts, and the Publish block can post to it. TikTok, Instagram, YouTube and X are supported, and one agent can rotate across several accounts on the same platform.

A video lands in one of three places, and the block says which before you run: saved as a draft in Overclip, sent to the platform drafts for you to post yourself, or published publicly straight away. Only TikTok and YouTube accept drafts.

Nothing is ever posted without an explicit run or a schedule you set, and disconnecting an account revokes our access to it.

Overclip exposes an MCP server, so an assistant that speaks the protocol can list your agents, estimate what a run would cost, launch it and read the result. Create a key in Settings, point your client at the server URL, and ask it in plain language.

Keys carry their own scopes and their own daily credit ceiling, which is checked on every run. An assistant cannot spend more than you allowed it, and a run it starts publishes only if the agent itself has a Publish block set to do so.