Runway Characters knowledge base: 50,000 tokens per avatar vs Sume

Runway's Characters docs say each avatar supports up to 50,000 tokens of knowledge, in plain text or Markdown, attached through documentIds. Sume avatars have no knowledge base. The avatar says exactly the script or video_inputs text you send, within a 4 to 60 second window per job.

What Runway documents

From the Characters knowledge base page, read 2026-10-02:

• The page says Runway plans to raise the token limit later.
• It does not state a maximum number of documents or a file size limit.

Why Sume has no equivalent

A Runway Character retrieves from documents during a live conversation. Sume's avatar video is a render: it takes the finished words, speaks them and returns a file. Nothing is looked up at playback time, so there is nothing for a knowledge base to do.

Using your own documents with Sume

If your content lives in documents, the retrieval and writing happen before the render. Draft the script from your document in your own code or in an agent, check it against the 4 to 60 second estimate, then submit. A script that is too long should be shortened or split into more jobs.

Because the script is the only content, review it before spending anything. Avatar previews let you approve the first frame before the full render.

• Document in, short script out, then render.
• One idea per clip keeps you under a minute.
• Keep the source document linked in your own records so a claim can be traced.

Choosing

Pick a Runway Character when users ask open questions and the answers come from your documents. Pick Sume when the message is fixed and you want a reviewable video file.