Video generation

Models

A value a model doesn't support is refused with a 400 that lists what it does.

Start a video

curl https://boldrouter.com/v1/videos \
  -H "Authorization: Bearer $BOLD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/gemini-omni-1.1-flash",
    "prompt": "A slow walk through a bright Scandinavian living room, single continuous shot, no dialogue",
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "seconds": 6
  }'

The response returns at once, with the video still generating:

{
  "id": "video_4bdf63d3fae6f69acd790312",
  "object": "video",
  "model": "google/gemini-omni-1.1-flash",
  "status": "in_progress",
  "created_at": 1793112000,
  "completed_at": null,
  "expires_at": null,
  "resolution": "720p",
  "aspect_ratio": "16:9",
  "size": "1280x720",
  "seconds": 6,
  "cost_usd": null,
  "content_url": null,
  "error": null
}

Check the status

Poll every few seconds until status is completed or failed. Once completed, seconds is the actual length, cost_usd what you were charged, and content_url where to download it.

import time, requests

H = {"Authorization": f"Bearer {API_KEY}"}
video = requests.post("https://boldrouter.com/v1/videos", headers=H, json={
    "model": "google/gemini-omni-1.1-flash",
    "prompt": "A paper boat drifting on a calm pond at sunrise",
}).json()

while video["status"] in ("queued", "in_progress"):
    time.sleep(5)
    video = requests.get(f"https://boldrouter.com/v1/videos/{video['id']}", headers=H).json()

if video["status"] == "completed":
    mp4 = requests.get(video["content_url"], headers=H).content
    open("boat.mp4", "wb").write(mp4)
else:
    print(video["error"]["message"])

Download the video

Returns the MP4 (H.264 with an audio track). Byte ranges are supported, so you can stream it to a player. Videos are kept for 30 days (expires_at); download the ones you want to keep.

Edit and extend a video

Edit one of your finished videos by describing the change, or continue it from its last frame. Each starts a new video, which you follow like any other; the original stays as it is.

curl https://boldrouter.com/v1/videos/video_4bdf63d3fae6f69acd790312/extend \
  -H "Authorization: Bearer $BOLD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "prompt": "The camera pulls back to reveal the whole lake", "seconds": 6 }'

• The result keeps the source's shape and resolution. An extension returns the whole video: the original followed by the new seconds.
• Gemini Omni continues the job that made the video, so nothing is uploaded and an edit changes only what you describe. It bills per token as usual, plus about $0.003 per second of the source it reads.
• The response says what was done: operation is edit or extend, and source_video_id the video it came from.

To edit or extend a video you didn't make here, send it as an MP4 (up to 20 MB, with Grok) to POST /v1/videos/edits or POST /v1/videos/extensions, with the same fields plus "video": "data:video/mp4;base64,...".

List and delete

• GET /v1/videos lists your videos, newest first (limit, up to 100, and after for the next page).
• DELETE /v1/videos/{id} deletes a finished video and its file.
• A workspace member's key sees only the videos that member made.

Pricing

You pay what the provider bills. When you start a video, the most it could cost is held from your balance, and released when it finishes.

A Grok edit or extension bills the seconds it makes (the whole video for an edit, the added seconds for an extension) plus about $0.01 per second of the source: editing a 4 second 480p video costs about $0.24. A Gemini Omni edit or extension bills like a new video of the seconds it makes, plus the source it reads: about $0.11 for 3 seconds at 360p.

Gemini Omni Flash bills per token: the prompt, the model's thinking, and the video itself. That's about $0.035 per second at 360p and about $0.10 per second at 720p, so a 6 second 720p clip costs around $0.60.

Good to know

• Prompts and images that break the provider's usage policies are refused with a 400 saying so, for example images of recognizable people, or of minors in the EU, Switzerland and the UK.
• With a starting image, Grok keeps the image's shape; aspect_ratio and size then describe what was asked for, not the file.
• Omni videos carry Google's invisible SynthID watermark. Prompts in English work best.
• By default Omni may cut between a few shots. Ask for "a single continuous shot" if you want one, and say "no dialogue" or describe the sound you want.
• Prompts can be anonymized like chat messages: send "anonymize": true. For workspace members with data protection on it always happens. Images aren't checked for private information.