Multi-shot video generation: what it is and how to do it

Multi-shot video generation means one request returns a single clip that already contains several shots, with cuts between them, instead of one continuous take. Some video APIs build it in, such as Kling 3.0 and Runway's Multi-Shot Video recipe. On an API without it, you generate each shot as its own clip and join the clips in order.

The Kling and Runway facts come from their own API docs, read on 2026-09-28. The Sume facts come from the Video generation and Timeline 1.0 docs.

Which video APIs support multi-shot generation?

Kling's capability map lists multi-shot generation for Kling 3.0 Turbo, 3.0 and 3.0 Omni, and not for Kling O1, 2.6 or 2.5 Turbo. Runway offers it as a recipe. Sume's video endpoint has no multi-shot option:

What is multi-shot in Kling?

In Kling's 3.0 Omni text-to-video API you write the shot list into the prompt as "shot n, m, words;", separated by semicolons: n is the shot number (1–6), m its length in seconds (at least 1, and all shots must add up to the video's duration), and words that shot's prompt, up to 512 characters. settings.multi_shot defaults to true; set to false, a multi-shot prompt does not produce multi-shot output.

Sume's kling-3 takes a prompt and a duration, but Sume's docs don't describe what a shot-list prompt does there, so don't build on it.

How do I make a multi-shot video on Sume?

Build it shot by shot, then cut it together:

• Write a shot list: one prompt and one length per shot. A shot list template helps.
• Generate each shot with POST /v1/videos. To keep one character across shots, send the same image as frame_images (first_frame) or input_references; Consistent character across shots compares the two.
• Join the clips in order with POST /v1/timeline-1.0/render: 1–200 video[] slots, the first starting at 0 and each later start increasing. Leave out transition for a hard cut, or add fade, wipeleft and others up to 1 second.
• Timeline reads only this workspace's media.sume.com files, such as the clips you just generated. Its default output is 1080×1920, so set output.width and output.height for a landscape sequence.
• Plan the sound. In current code a Timeline render takes sound only from its audio spine and an optional soundtrack, and drops each clip's own audio. Send a voiceover as audio.url, or audio.mode: "silence" for none.

curl -X POST https://api.sume.com/v1/timeline-1.0/render \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: multishot-001" \
  -d '{
    "audio": { "mode": "silence", "duration_seconds": 15 },
    "video": [
      { "source_url": "https://media.sume.com/artifacts/artf_demo/shot1.mp4", "start": 0, "duration": 5 },
      { "source_url": "https://media.sume.com/artifacts/artf_demo/shot2.mp4", "start": 5, "duration": 5 },
      { "source_url": "https://media.sume.com/artifacts/artf_demo/shot3.mp4", "start": 10, "duration": 5 }
    ]
  }'

Should I use one multi-shot request or stitch shots?

They trade off differently. A built-in multi-shot request is one generation, but the whole sequence fits inside one clip's length (15 seconds on Kling 3.0 and on Runway's recipe), and changing one shot means generating the whole clip again. Stitching costs one request per shot plus the render, listed at $0.10 per output minute on API pricing, but you can redo one shot, mix models, and run the spine up to 1,800 seconds. How to assemble a long-form video covers the Timeline request in full.