Input files

Five API takes input files as URLs. There is no upload endpoint: host the file somewhere public, for example in your own object storage, and pass its HTTPS URL in the request.

Field	Used by	What it points to
`image_url`	image to video	The image to animate
`reference_images`, `reference_videos`, `reference_audios`	reference to video	Arrays of asset URLs
`video_url`	video extend	The clip to continue

URL policy

Five API checks every input URL when you create the task, before any credit is held. A URL that breaks a rule answers 400 url_not_allowed with the field in param.

HTTPS only. http:// URLs are rejected.
Public host. The host name must resolve, through public DNS, to a public IP address. Private, loopback, link-local, carrier-grade NAT, cloud metadata and other reserved ranges are rejected.
Answers HEAD. Five API sends a HEAD request to the URL. It must answer 2xx, after at most 3 redirects; each redirect target is checked again.
Within the size limit. The Content-Length must not exceed the limit of the model (see below).

The same host rules apply to webhook_url.

Presigned URLs from S3, Google Cloud Storage, Azure Blob Storage, Cloudflare R2 and similar services work well. Keep them valid for at least an hour: a task downloads its input files after it is created, while it is queued.

Limits per model

Model	Input	Formats and limits
Video Fast 1.0, image to video	`image_url`	JPG, PNG or WebP; shorter side at least 300 px; aspect ratio between 0.4 and 2.5; at most 30 MB
Wan 3.0 Spark and Prime Spark, image to video	`image_url`	JPG, PNG or WebP; 360–2,000 px per side; at most 20 MB
Video Fast 1.0, reference to video	`reference_images`	Up to 9 images: JPG, PNG or WebP; 256–5,760 px; at most 30 MB each
	`reference_videos`	Up to 3 clips: 2–15 seconds and at most 50 MB each
	`reference_audios`	Up to 3 files: at most 15 MB each; send them together with an image or a video
Wan 3.0 Spark, reference to video	`reference_images`	Up to 10 images: JPG, PNG or WebP; at most 20 MB each
	`reference_videos`	Up to 5 clips: at most 100 MB each
	`reference_audios`	Up to 5 files: at most 15 MB each
Video Fast Extend 1.1	`video_url`	MP4 or MOV; 360–2,000 px per side; 2–30 seconds after the optional window; at most 500 MB

GET /v1/models/{brand}/{version}/{task} returns each model’s limits in input_schema, including x-max-input-bytes, the largest accepted file.

When an input fails later

The URL check at creation can’t see everything inside a file. If a file turns out to be unusable during processing, the task fails and its credits are released:

input_fetch_failed (retryable): the file could not be downloaded or processed, for example because a presigned URL expired.
input_invalid (not retryable): the file was downloaded but its format, dimensions or length do not fit the model.

Troubleshooting `url_not_allowed`

Cause	Fix
The URL uses `http`	Serve it over `https`
The host resolves to a private or reserved address	Use a publicly hosted file; internal hosts and DNS overrides are not reachable
The server doesn’t answer `HEAD`, or answers without `Content-Length`	Host the file on object storage or a CDN that supports `HEAD`
More than 3 redirects	Pass the final URL
The file is larger than the model’s limit	Compress or resize it, or choose a model with a higher limit

Get started

Core concepts

Guides

Build with AI

Resources

URL policy

Limits per model

When an input fails later

Troubleshooting `url_not_allowed`

​URL policy

​Limits per model

​When an input fails later

​Troubleshooting url_not_allowed

URL policy

Limits per model

When an input fails later

Troubleshooting `url_not_allowed`