Multimodal

Motion control

Kling v3.0 Motion Control — a reference video's motion and expression puppeteers a character photo. Unlike Lip sync and Face swap, this is a real POST /v1/videos request shape, not a separate endpoint — both inputs are a character image (start_image_url) plus a reference motion video, just carried in a different field depending on which of the two orientations you want.

Swap Avatar (image orientation)

The reference motion video goes in input_video_references (exactly one entry) alongside start_image_url — the one combination on this platform where a reference array and a starting frame are both allowed at once. Output is a fixed ~4 seconds, following the photo's own composition rather than the reference video's length.

import requests

job = requests.post(
    "https://videorouter.sh/api/v1/videos",
    headers={"Authorization": "Bearer llmr_sk_live_..."},
    json={
        "model": "kling/motion-control-std/novita",
        "start_image_url": "https://example.com/character.jpg",
        "input_video_references": [
            {"type": "video_url", "video_url": {"url": "https://example.com/reference-motion.mp4"}}
        ],
    },
).json()
# -> {"id": "video_...", "status": "queued", ...}

curl https://videorouter.sh/api/v1/videos \
  -H "Authorization: Bearer llmr_sk_live_..." \
  -H "content-type: application/json" \
  -d '{
    "model": "kling/motion-control-std/novita",
    "start_image_url": "https://example.com/character.jpg",
    "input_video_references": [
      {"type": "video_url", "video_url": {"url": "https://example.com/reference-motion.mp4"}}
    ]
  }'
# -> {"id": "video_...", "status": "queued", ...}

Act Mimic (video orientation)

The reference motion video instead goes in input_video_url (the ordinary video-editing field), still paired with start_image_url — unlike every other editing model on this platform, neither field requires a prompt. Output duration matches the reference video's own probed length, up to 30 seconds.

import requests

job = requests.post(
    "https://videorouter.sh/api/v1/videos",
    headers={"Authorization": "Bearer llmr_sk_live_..."},
    json={
        "model": "kling/motion-control-pro/novita",
        "start_image_url": "https://example.com/character.jpg",
        "input_video_url": "https://example.com/reference-motion.mp4",
    },
).json()
# -> {"id": "video_...", "status": "queued", ...}

curl https://videorouter.sh/api/v1/videos \
  -H "Authorization: Bearer llmr_sk_live_..." \
  -H "content-type: application/json" \
  -d '{
    "model": "kling/motion-control-pro/novita",
    "start_image_url": "https://example.com/character.jpg",
    "input_video_url": "https://example.com/reference-motion.mp4"
  }'
# -> {"id": "video_...", "status": "queued", ...}

Both orientations bill the same $/s rate regardless of which field carries the reference video — that cost already includes our 2% platform fee. Poll with the ordinary GET /v1/videos/{id}.

Which models

One provider today — no cross-provider failover pool yet. Also shows up under the Motion Control filter on the models catalog.