venice-video

Venice Video

Video is asynchronous — like audio music. Five endpoints:

Use when

• You need text-to-video, image-to-video, video upscale, video-with-audio, or video transcription.
• You can tolerate async execution (single-digit seconds to several minutes depending on model, duration, and queue depth — inspect average_execution_time and execution_duration on /video/retrieve for your job's live estimate).
• You want to price a job precisely before committing (/video/quote).

Lifecycle — generation

1. Price with /video/quote

curl https://api.venice.ai/api/v1/video/quote \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan-2-7-text-to-video",
    "duration": "5s",
    "aspect_ratio": "16:9",
    "resolution": "720p",
    "audio": true
  }'

Response: {"quote": 0.35} USD.

2. Submit with /video/queue

curl https://api.venice.ai/api/v1/video/queue \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "wan-2-7-text-to-video",
    "prompt": "Commerce being conducted in the city of Venice, Italy.",
    "negative_prompt": "low resolution, worst quality, defects",
    "duration": "5s",
    "aspect_ratio": "16:9",
    "resolution": "720p",
    "audio": true
  }'

Response: { "model": "...", "queue_id": "uuid", "download_url": "https://..." }.

• download_url only appears for VPS-backed models. When present, the retrieve endpoint returns JSON status only — fetch this URL to download. Valid 24 h.

3. Poll with /video/retrieve

curl https://api.venice.ai/api/v1/video/retrieve \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"...","queue_id":"..."}' \
  --output out.mp4

• Processing: JSON {"status":"PROCESSING","average_execution_time":145000,"execution_duration":53200} (ms).
• Completed (non-VPS): binary video/mp4 body.
• Completed (VPS-backed): {"status":"COMPLETED", ...} — fetch the download_url from the queue response.
• delete_media_on_completion: true auto-deletes after successful retrieve.

4. Finalize with /video/complete

curl https://api.venice.ai/api/v1/video/complete \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"...","queue_id":"..."}'

QueueVideoRequest fields

Availability depends on the model — check GET /models?type=video.

Common recipes

Text → video with audio

{
  "model": "wan-2-7-text-to-video",
  "prompt": "A golden retriever chasing a frisbee in slow motion at sunset.",
  "duration": "6s",
  "aspect_ratio": "16:9",
  "resolution": "720p",
  "audio": true
}

Image → video

{
  "model": "<image-to-video model>",
  "prompt": "Camera slowly zooms out, revealing the cityscape.",
  "image_url": "https://example.com/cityscape.jpg",
  "duration": "5s",
  "aspect_ratio": "16:9"
}

Video upscale

{
  "model": "<upscale model>",
  "video_url": "data:video/mp4;base64,...",
  "upscale_factor": 2,
  "duration": "Auto"
}

Multi-element consistency (Kling O3 R2V-style)

{
  "model": "<advanced-model>",
  "prompt": "@Element1 walks toward @Element2 against @Image1.",
  "elements": [
    { "frontal_image_url": "<char1.png>", "reference_image_urls": ["<alt1.png>"] },
    { "frontal_image_url": "<char2.png>" }
  ],
  "scene_image_urls": ["<street-scene.jpg>"]
}

/video/transcriptions (sync)

Transcribe a YouTube video URL directly — no queue.

curl https://api.venice.ai/api/v1/video/transcriptions \
  -H "Authorization: Bearer $VENICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url":"https://www.youtube.com/watch?v=...","response_format":"json"}'

Response: {"transcript":"...","lang":"en"} (JSON) or plain text/plain body when response_format: text.

For arbitrary audio files, use venice-audio-transcription instead.

Full polling loop

async function waitForVideo(model: string, queueId: string, downloadUrl?: string) {
  while (true) {
    const res = await fetch(`${base}/video/retrieve`, {
      method: 'POST', headers,
      body: JSON.stringify({ model, queue_id: queueId }),
    })
    const ct = res.headers.get('content-type') ?? ''
    if (ct.startsWith('video/')) {
      return Buffer.from(await res.arrayBuffer())
    }
    const body = await res.json()
    if (body.status === 'COMPLETED' && downloadUrl) {
      const v = await fetch(downloadUrl)
      return Buffer.from(await v.arrayBuffer())
    }
    if (body.status !== 'PROCESSING') throw new Error(`unexpected ${body.status}`)
    await new Promise(r => setTimeout(r, 5000))
  }
}

Errors

/video/queue does not document 503 in the spec — upstream capacity issues surface there as 500. Watch for 503 specifically on /video/retrieve.

Venice Studio Web UI (Not Self-Hosted)

Venice Studio is browser-based at venice.ai/studio — it is NOT a local package or self-hosted app. It provides a 6-tab workspace: Image, Edit, Audio, Video, Movie Editor, and Library. The Movie Editor supports timeline-based editing, transitions, audio layers, and title cards that the Venice API does not expose.

Implication: The Venice API (and therefore Hermes) is ideal for bulk generation and iteration — treatments, character refs, shot lists, raw clips, music tracks. The Venice Studio web UI is required for final assembly — timeline editing, transitions, audio sync, and Seedance 2.0 cinematic shots. Plan a hybrid workflow: generate assets via API/Hermes, then assemble and polish in the Studio browser UI.

See references/venice-studio-vs-api.md for a full feature matrix and workflow diagrams.

Gotchas

• duration is required on /video/queue. Even Auto is a valid explicit value.
• download_url is only sometimes returned at queue time. Always handle both paths: binary from /retrieve OR fetching download_url after status COMPLETED.
• download_url expires in 24 h — download promptly.
• Upscale models use upscale_factor instead of resolution.
• reference_image_urls[] is capped at 9 entries, elements[] at 4, scene_image_urls[] at 4. Over-limit is 400.
• data: URLs count toward payload size; large base64 videos may trip 413 — prefer hosted URLs.
• /video/transcriptions is YouTube-URL-only; it does not accept arbitrary video uploads (use ffmpeg to strip audio, then /audio/transcriptions).
• Seedance 2.0 is UI-only as of mid-2025. If a user asks for "the best cinematic quality" or "Seedance," route them to venice.ai/studio — the API cannot reach it yet.

Audio parameter model inconsistency

/video/quote may accept audio: true for a model, but /video/queue can still reject it with:

{"details":{"audio":{"_errors":["This model does not support audio configuration"]}},"issues":[{"code":"custom","message":"This model does not support audio configuration","path":["audio"]}]}

Sora 2 Pro is a known silent-only model — omit audio entirely when queueing to it. Always test-quote first, then if queue fails on audio, retry without the audio field. See references/model-audio-matrix.md for a tested compatibility list.

/models?type=video endpoint quirks

GET /models?type=video returns video models but does not populate model_spec.inference_type — the field is null for all entries. You must check model_spec.constraints.durations[] and model_spec.capabilities to distinguish video from text models. The type query parameter works (?type=video gives 92 models, ?type=text gives 75), but the JSON schema lacks the usual inference-type hints.