Video

Use POST /v1/video/generations for synchronous video generation. The HTTP connection stays open while the provider renders the result, so set a generous client timeout.

POST https://api.rout.my/v1/video/generations

Available models

Model	What it does	Resolutions	Duration	Audio
bytedance/seedance-2.5	Text-to-video, image-to-video, first/last frame, reference-to-video, video editing, extend	480p, 720p	4–30 s	Yes
qwen/wan-v2.7	Text-to-video and reference-to-video (mode selected automatically)	720p, 1080p	2–15 s (t2v), 2–10 s (r2v)	Yes

google/gemini-omni-flash-preview can also emit video, but it is a language model on /v1/chat/completions, not a video endpoint. It is not accepted here.

Always use the exact model IDs returned by /v1/models.

Request

curl --max-time 900 https://api.rout.my/v1/video/generations \
  -H "Authorization: Bearer ***" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bytedance/seedance-2.5",
    "prompt": "A slow camera move across a glass cube on a white table",
    "duration": 5,
    "resolution": "720p",
    "aspect_ratio": "16:9"
  }'

Request fields

Field	Type	Required	Notes
model	string	Yes	Exact video model ID from /v1/models.
prompt	string or object	Yes	Text prompt, or { "text": "...", "image": "..." } for image-to-video.
duration	number	No	Seconds. Clamped to the model's range.
resolution	string	No	480p, 720p, 1080p, or WIDTHxHEIGHT depending on the model.
aspect_ratio, aspectRatio	string	No	16:9, 9:16, 1:1, 4:3, 3:4, 21:9. Ignored when an input image defines it — see below.
n, sampleCount	integer	No	Number of outputs. Seedance always returns exactly 1.
audio	boolean	No	Enable generated audio where supported.
negative_prompt, negativePrompt	string	No	Content to avoid.
seed	integer	No	Deterministic seed when supported.
input_image	string	No	URL, data URI, or raw base64. Used as the first frame.
image, image_url, input_image_url, input_image_b64	string	No	Aliases for the image input.
last_frame_image, lastFrameImage, image_tail, imageTail	string	No	Ending frame for first/last-frame generation.
reference_images, referenceImages	string[]	No	Image references that guide content without pinning the first frame.
reference_videos, referenceVideos	string[]	No	Video references.
reference_urls, referenceUrls	string[]	No	Mixed references; video items are inferred from the file extension.
input_video, input_video_b64	string	No	Source clip for editing/extending. Triggers video-input pricing.
input_audio, input_audio_b64, audio_url, audioUrl	string	No	External audio track where supported.
prompt_enhancement, enhance_prompt, prompt_extend, enable_prompt_expansion	boolean	No	Prompt expansion switch.
camera_fixed, cameraFixed	boolean	No	Lock the camera (Seedance).
shot_type, shotType	string	No	Wan reference-to-video shot type: single or multi.
watermark	boolean	No	Watermark flag where supported.
service_tier, serviceTier	string	No	default or flex on Seedance.
poll_timeout_ms, poll_interval_ms	integer	No	Job polling timeout and interval.
user	string	No	Client-side user identifier.

Base64 and data URI media are temporarily hosted as URLs when the provider requires a URL.

Seedance 2.5

One model covers six modes; which one runs is inferred from the fields you send.

Mode	Send
Text-to-video	prompt only
Image-to-video	input_image, or prompt: { text, image }
First/last frame	input_image + last_frame_image
Reference-to-video	reference_images and/or reference_videos
Video editing	input_video + a prompt describing the change
Extend video	input_video with a continuation prompt

Limits: 480p or 720p, 4–30 seconds, 24 fps, exactly one video per call. Input images: up to 30, JPEG/PNG/WebP/BMP/TIFF/GIF/HEIC/HEIF. Input videos: up to 10 clips, MP4/MOV, 2–30 s each. Input audio: MP3/WAV, 2–30 s.

Aspect ratio and input images

When you send a first-frame image, do not send aspect_ratio. ByteDance derives the output ratio from that image and rejects the request outright if a ratio is also specified:

The parameter ratio specified in the request is not valid. For first-frame
or first-last-frame generation, the output ratio follows the first-frame image.

rout.my strips aspect_ratio for you in this case, so a request carrying both still succeeds — the image wins. Crop the image to the ratio you want.

aspect_ratio does apply to text-to-video and to reference-to-video, since neither pins the frame.

{
  "model": "bytedance/seedance-2.5",
  "prompt": {
    "text": "Animate this product with a slow turntable movement.",
    "image": "https://example.com/product.png"
  },
  "duration": 5,
  "resolution": "720p"
}

References

Reference images guide characters, style, or props without becoming the first frame. Refer to them positionally in the prompt.

{
  "model": "bytedance/seedance-2.5",
  "prompt": "Use [Image 1] as the main character and [Image 2] as the product in a cinematic street scene.",
  "reference_images": [
    "https://example.com/character.png",
    "https://example.com/product.png"
  ],
  "duration": 5,
  "resolution": "720p",
  "aspect_ratio": "16:9"
}

When input_image is combined with audio or video references, rout.my routes the image to reference_images instead of the first frame, so reference media and the frame do not conflict.

Editing and extending

{
  "model": "bytedance/seedance-2.5",
  "prompt": "Make the camera movement smoother and keep the same subject.",
  "input_video": "https://example.com/source.mp4",
  "duration": 5,
  "resolution": "720p"
}

A request whose input contains video is billed at a reduced rate, but the source clip's own duration is billed too — see below.

Wan 2.7

qwen/wan-v2.7 routes to the right upstream mode automatically:

• no reference fields → text-to-video (720p/1080p, 2–15 s)
• any reference_images, reference_videos, or reference_urls → reference-to-video (2–10 s, up to 5 images or 3 clips)

{
  "model": "qwen/wan-v2.7",
  "prompt": "Image 1 and Video 1 walk through a rainy city",
  "duration": 6,
  "resolution": "1080p",
  "reference_images": ["https://example.com/character.png"],
  "reference_videos": ["https://example.com/motion.mp4"]
}

Prefer the typed reference_images / reference_videos fields when mixing media, so the router does not have to guess from filenames. Wan accepts resolution as 720p/1080p or as WIDTHxHEIGHT.

Response

{
  "created": 1744000000,
  "model": "bytedance/seedance-2.5",
  "duration": 5,
  "videos": [
    {
      "url": "https://api.rout.my/public/media/generated.mp4",
      "mime_type": "video/mp4"
    }
  ]
}

Generated URLs are temporary. Download and re-host anything you need to keep.

Quota accounting

Two models of billing are in use, depending on the provider.

Seedance 2.5 is billed per video token, matching ByteDance's own metering:

tokens = (output_sec + input_video_sec) x width x height x 24 / 1024

Both resolutions bill at the same rate per token, so the price difference comes from the pixel count. A minimum token floor applies per clip based on output duration, and any request whose input contains video is billed at the reduced rate — including the input clip's own seconds.

Resolution	Tokens per second	Cost per second
480p (854x480)	9,607.5	$0.1028
720p (1280x720)	21,600	$0.2311

With video in the input the rate drops from $10.70 to $6.40 per million tokens, i.e. $0.0615/s at 480p and $0.1382/s at 720p.

Wan 2.7 is billed per second of output: $0.10/s at 720p and $0.15/s at 1080p.

Quota tokens are derived from the same figures and scaled by the model's token_multiplier.