Google Video API

This page specifies the Google video models veo-3.1, veo-3.1-fast, gemini-omni-flash and gemini-omni-flash-1.1. They share the POST /v1/videos create and GET /v1/videos/{id} polling flow with the other video models; authentication, media upload, polling states, downloads and error handling are covered in the Complete Video API Standard. This page lists only the fields and rules specific to Google video.

Updated 2026-10-10. New Google video requests use polling; do not send callback_url.

Models

Common rules

• Aspect ratio: ratio (or aspect_ratio) accepts only 16:9 and 9:16. The default is 16:9.
• Resolution: resolution accepts 720p, 1080p or 4k. The default is 720p. A 360p request is generated and billed as 720P.
• size: one field can set both resolution and ratio, for example 1280x720, 1080x1920 or 3840x2160. It can also carry only the ratio, 16:9 or 9:16. Conflicting values in size, ratio and resolution return 400.
• Count: each request generates one video, so n must be 1.
• Images: up to 20 MiB each, in any one of these forms:
  • a public HTTPS URL (recommended);
  • a Data URI, for example data:image/png;base64,...;
  • raw PNG, JPEG or WebP Base64 without a prefix;
  • a multipart/form-data file in the field input_reference, image, images, first_frame or last_frame.
• Image formats: Veo accepts JPEG and PNG only. Gemini Omni accepts PNG, JPEG and WebP.
• Videos: public HTTPS URLs only, MP4, MOV or WebM, up to 64 MiB. Upload local videos first (see Local media: permit, upload, reference) and put the returned url in video. Video Data URIs, Base64 and multipart video files are not accepted.
• Reference video field: use top-level video or videos, not metadata.referenceVideos. No Google video model accepts reference audio.
• Result: on completion, video_url is an MP4 on assets.88api.ai. Download it directly, or use it as the video of a later extension or edit request.

Veo 3.1 / Veo 3.1 Fast

• metadata.video_mode selects frames (first frame or first and last frames) or reference (reference images) explicitly. task can also be text_to_video, image_to_video, reference_to_video or extend. Conflicting values return 400.
• generate_audio (or metadata.generateAudio) defaults to true and generates native synchronized audio. With false, the output has no audio track. Audio does not change the price.
• negative_prompt (up to 2000 characters) and seed (0–4294967295) can be top-level or written as metadata.negativePrompt and metadata.seed.
• The extension input must be a 1–30 second MP4. Extensions are billed by added seconds: output duration minus input duration.
• Veo does not accept reference videos, reference audio, temperature or top_p.

4K text to video:

{
  "model": "veo-3.1",
  "prompt": "A silver sports car in a rainy city at night, low tracking shot, neon reflections, cinematic",
  "duration": 8,
  "resolution": "4k",
  "ratio": "16:9",
  "generate_audio": true
}

First and last frames (the two images are the first and last frame, in order):

{
  "model": "veo-3.1-fast",
  "prompt": "Transition smoothly from an empty square at dawn to the same view at night with the lights on",
  "duration": 6,
  "size": "1920x1080",
  "images": [
    "https://cdn.example.com/start.jpg",
    "https://cdn.example.com/end.jpg"
  ],
  "metadata": {
    "negativePrompt": "blur, distortion"
  }
}

Reference images (fixed 8s):

{
  "model": "veo-3.1",
  "prompt": "Keep the face and outfit from the reference images and walk slowly along the beach",
  "duration": 8,
  "images": [
    "https://cdn.example.com/character-front.jpg",
    "https://cdn.example.com/character-side.jpg"
  ],
  "metadata": {
    "video_mode": "reference"
  }
}

Video extension:

{
  "model": "veo-3.1",
  "prompt": "Continue the shot as the person keeps walking forward",
  "task": "extend",
  "video": "https://assets.88api.ai/media/task-videos/2026/10/example.mp4"
}

Upload a local image as the first frame (metadata is sent as a JSON string):

curl --request POST 'https://api.88api.ai/v1/videos' \
  --header 'Authorization: Bearer sk-xxxx' \
  --form 'model=veo-3.1-fast' \
  --form 'prompt=Slowly push in while the clouds drift in the background' \
  --form 'seconds=8' \
  --form 'size=1280x720' \
  --form 'metadata={"video_mode":"frames"}' \
  --form 'input_reference=@start.jpg;type=image/jpeg'

Gemini Omni Flash / Gemini Omni 1.1

Without task, the submitted media decide the task:

• task can also be set explicitly to text_to_video, image_to_video, reference_to_video, edit or extend.
• Unlike Veo, two images in images are treated as references. For first and last frames, use first_frame and last_frame.
• First/last frame fields cannot be combined with ordinary reference images.
• gemini-omni-flash outputs 720P only and does not support editing or extension. Submit its reference video together with reference images.
• gemini-omni-flash-1.1 accepts up to 10 reference images and 3 reference videos.
• Video editing (1.1): put one video in video and describe the change in the prompt. Do not send ratio, aspect_ratio or size; the aspect ratio follows the source video. resolution may be set.
• Video extension (1.1):
  • Send task: "extend" with one video. Do not send duration, ratio or size.
  • Each call adds about 10 seconds. resolution can be 720p (default) or 1080p; 4K is not supported.
  • The input must be a 1–30 second MP4. For a longer video, cut the last part (10 seconds recommended) and extend that.
  • Billing uses added seconds: output duration minus input duration. Optionally send input_duration (input length in seconds); a value more than 0.1 seconds away from the measured length is rejected.
• The extension result is a complete MP4. Its input part is regenerated by the model: similar to the original, not a frame-exact copy. To stitch, keep the original in full and append only the part of the extension after the input duration.
• temperature (0–2) and top_p (0–1) are optional. Gemini Omni has no audio switch.

First frame (Flash):

{
  "model": "gemini-omni-flash",
  "prompt": "Slowly push in as steam rises from the pastries",
  "duration": 5,
  "ratio": "9:16",
  "image": "https://cdn.example.com/start.jpg"
}

Reference images plus reference video (1.1, 4K):

{
  "model": "gemini-omni-flash-1.1",
  "prompt": "Keep the camera rhythm of the reference video and replace the subject with the product from the image",
  "duration": 8,
  "resolution": "4k",
  "images": ["https://cdn.example.com/product.jpg"],
  "video": "https://cdn.example.com/reference.mp4"
}

Video extension (1.1):

{
  "model": "gemini-omni-flash-1.1",
  "prompt": "Continue the scene as ripples spread naturally across the water",
  "task": "extend",
  "resolution": "1080p",
  "video": "https://assets.88api.ai/media/task-videos/2026/10/example.mp4"
}