Grok Imagine video extension vs chaining clips on Sume

xAI's video guide documents an extension call that continues a Grok Imagine clip from its final frame. Sume does not expose that call for Grok. On Sume you get the same result by hand: save the last frame of the finished clip and submit it as the first_frame of the next grok-imagine-video-1.5 job. This post shows the difference and the chain.

What each side documents

The table lists what each side documents. xAI facts come from its video generation guide; Sume facts come from the Video Router and Video Generation docs.

Chain two clips on Sume

A chain has three steps. Finish clip one and download it. Extract its last frame. Submit clip two with that frame in frame_images and frame_type: "first_frame". Sume treats frame_images as image-to-video, which is the only mode the Grok row accepts.

Each hop is a separate Sume job with its own id, price and poll. That is the main difference from a native extension: you pay and poll per hop, and you decide how long each segment is (4-15 s on Sume).

Quality drift is yours to watch. A single still carries no motion information, so describe the motion that should continue in the prompt of each hop.

ffmpeg -sseof -0.1 -i clip1.mp4 -frames:v 1 last.png
# upload last.png to a public https URL, then:
curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "grok-imagine-video-1.5",
    "prompt": "The runner keeps sprinting down the track, camera follows",
    "duration": 8,
    "resolution": "720p",
    "frame_images": [{
      "type": "image_url",
      "image_url": {"url": "https://example.com/last.png"},
      "frame_type": "first_frame"
    }]
  }'

Which to use

Use xAI's own extension when you call xAI directly and want the continuation handled for you. Use the chain when you are on Sume and want one billing and one API for every model. If you need a longer single take, check whether another catalog row such as seedance-2.5 (4-30 s) fits before you chain.

Check the live catalog with GET /v1/videos/models before you hard-code a duration range.