Multimodal

Video upscaling

Same POST /v1/videos endpoint as generation, using the same input_video_url field video editing uses — pass an existing video and get back a higher-resolution version of that same clip. No prompt is required. There's no duration field either — output duration always equals your input video's own duration.

import requests, time

job = requests.post(
    "https://videorouter.sh/api/v1/videos",
    headers={"Authorization": "Bearer llmr_sk_live_..."},
    json={
        "model": "bytedance/seedvr2-upscale-video/fal",
        "scale_factor": 2,
        "input_video_url": "https://example.com/clip.mp4",
    },
).json()
# -> {"id": "video_...", "status": "queued", ...}

status = requests.get(
    f"https://videorouter.sh/api/v1/videos/{job['id']}",
    headers={"Authorization": "Bearer llmr_sk_live_..."},
).json()

const createRes = await fetch("https://videorouter.sh/api/v1/videos", {
  method: "POST",
  headers: { Authorization: "Bearer llmr_sk_live_...", "Content-Type": "application/json" },
  body: JSON.stringify({
    model: "bytedance/seedvr2-upscale-video/fal",
    scale_factor: 2,
    input_video_url: "https://example.com/clip.mp4",
  }),
});
const job = await createRes.json();
// -> {"id": "video_...", "status": "queued", ...}

const statusRes = await fetch(`https://videorouter.sh/api/v1/videos/${job.id}`, {
  headers: { Authorization: "Bearer llmr_sk_live_..." },
});

curl https://videorouter.sh/api/v1/videos \
  -H "Authorization: Bearer llmr_sk_live_..." \
  -H "content-type: application/json" \
  -d '{
    "model": "bytedance/seedvr2-upscale-video/fal",
    "scale_factor": 2,
    "input_video_url": "https://example.com/clip.mp4"
  }'
# -> {"id": "video_...", "status": "queued", ...}

curl https://videorouter.sh/api/v1/videos/video_... \
  -H "Authorization: Bearer llmr_sk_live_..."

Two request shapes depending on the model (see the table below): most fal.ai rows take scale_factor (a multiplier on your input video's own resolution, typically 1–4, default 2); the 3 Atlas Cloud rows instead take the existing resolution field (e.g. "1080p") since their own APIs expose an explicit output-resolution tier rather than a continuous factor. Real price depends on the OUTPUT resolution/duration, computed from your input video's own probed dimensions — that cost already includes our 2% platform fee.

Which models

11 models across 3 providers — fal.ai, Atlas Cloud, and Black Forest Labs (direct).

The two flux-video-upscale* rows call Black Forest Labs directly (flux-video-upscale is also a failover sibling of fal/flux-video-upscale). They accept a scale_factor of 1.5–3 (values outside are clamped), inputs up to 2560×1440 and 20 seconds, and settle from Black Forest Labs' own reported cost on the finished job.

The 3 Atlas Cloud rows bill through Atlas Cloud's reserve-then-settle flow (same as every other Atlas Cloud video model) — the price shown is a pre-call estimate; the real charge settles from Atlas Cloud's own completed-job invoice, same as every other Atlas Cloud video row. Every model here also shows up under the Video Upscale filter on the models catalog.

Not yet supported

Frame-rate interpolation (some providers support it upstream) isn't exposed as a request parameter. Topaz Astra 2 (fal/topaz-upscale-video-creative) auto-selects its own output resolution regardless of any request field — always billed at the conservative 4K ceiling. Atlas Cloud's Tencent row is fixed to its general-purpose "live-action-enhance" style; the anime/restore/small-face style variants aren't exposed. stream: true is rejected outright, same as generation.