Velo

View saved version →

---
name: velo-video
description: >
  This skill should be used when the user wants a video made with Velo, such as
  "make a video with Velo", "make an explainer / product demo / launch video",
  "turn this PDF (or deck, or screen recording) into a video", "make a video of this
  website", or "a video presented by an avatar". It also applies when following a
  Velo video ("check on my Velo video", "is my video ready?"), when Velo asks the
  user something (the brief, script approval, a voice or an avatar), when changing a
  draft, and when exporting or sharing a finished video.
---

# Making videos with Velo

Velo makes animated, narrated videos: explainers, product demos, training modules, pitch and launch videos, motion graphics and social clips. It works from a brief, a document, a screen recording, a public website or an AI presenter. Velo's own agent builds each video in a Velo chat; the Velo tools start it, relay the conversation and hand back the links.

The role here: write Velo a short brief that says what the video should achieve, show the user everything Velo says and asks, send back the user's answers and changes, and give them the links. Prompting Velo covers how to word what goes to Velo.

## How results read in chat

- Show every link Velo returns as a short markdown hyperlinks, such as [Open your video](url) or [Add your files](url). Never paste the raw URL, and never build or guess one.
- Keep IDs (`chat_id`, voice, avatar, option and scene IDs) out of chat. Use them only in tool calls.
- Use names and labels in every reply, confirmations included. Map the user's choices back to its ID when calling a tool.
- Say times and durations in plain words: "about 2 minutes", "an hour ago".
- Describe choices and next steps in everyday language.

## Start a video

1. **Add files first, if there are any.** A document, a recording, audio, images or a logo reaches Velo through `upload_file`.
   - Files the user attached in this conversation: pass them in `files`. Velo adds them directly and returns them in `files`. Any it couldn't take come back in `refused`, with the reason and an `upload_page_url` for adding them another way.
   - Any other file, such as one saved on the user's computer: call `upload_file` with no `files`. Give the user the returned `upload_page_url` as a link: they add the files in their browser and choose **Done uploading**. Wait until the user says they've uploaded the files, then call `list_files` with the returned `chat_id` to confirm the files arrived. It waits briefly while a file is still uploading.
   - To upload more files, call `upload_file` again with the same `chat_id` for a fresh link.
2. **Start it with `create_video`.** Write `message` as a short brief for Velo, 2 to 5 sentences, from the whole conversation:
   - Say what the video is about, what each attached file shows, and the conditions the user set (length, audience, tone, a script review first).
   - Describe files only by what they show ("the attached deck is our Q3 pitch"), never by where they came from or how they were fetched.
   - Shape it as "Prompting Velo" describes, from what the user said and gave. Leave out any script, scene list or facts dump of your own, and any detail the user left open. Velo's agent asks questions for what it needs.
   - If files were uploaded, pass that `chat_id`. The waiting files go with it. `files_left` says how many still wait for the next message; `files_refused` names any Velo couldn't take, with the reason.
3. **Pick the `flow`** that matches the request:

| flow | Use it for | Needs first |
| --- | --- | --- |
| `auto` | A topic, a brief, or a public website (put the URL in `message`; Velo opens it in its own browser, so no recording is needed). | A `message` |
| `upload_pdf` | A video from a document (PDF, deck). | The document, through `upload_file` |
| `upload_recording` | Polishing a screen recording. | The recording, through `upload_file` |
| `record_yourself` | A video of the user talk

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 1 — file references are metadata, not package contents.