Video generation

HopBase sells six video model families. Five run on one asynchronous task API — submit to POST /v1/video/generate, then poll the task ID it returns — while Wan 3.0 and HappyHorse keep Alibaba's own DashScope paths and Gemini Omni returns video synchronously.

All video models

Each family sits in its own plan group, so a key that reaches one does not reach the others. Gemini Omni video is the exception in one respect: it shares the Gemini group with Gemini chat and image, so no separate video package is needed there. Call GET /v1/models with the key you intend to use and read the complete model ID from the response before sending paid work. Model IDs are exact: no aliases or bare names are registered, and an unknown name is rejected rather than routed to a near match.

This page documents the Seedance models. Every other family has its own page — follow the links above.

Seedance 2.5 and 2.0 asynchronous task API

China and overseas variants

Existing clients moving from overseas to China only need to replace the group key. The base_url, API paths, request parameters, and polling logic remain unchanged. Use GET /v1/models with the new key to select an available complete model ID.

Seedance 2.5 uses model ID dreamina-seedance-2-5-260628 in the overseas group and doubao-seedance-2-5-260628-a in the China group (the China variant additionally supports 1080p). For ordinary text-to-video and reference-media generation, duration defaults to -1 and accepts integer values from 4-30 seconds or -1; resolution defaults to 720p and accepts 480p or 720p; and ratio accepts 16:9, 4:3, 1:1, 3:4, 9:16, 21:9, or adaptive. Some task modes have stricter requirements: video editing uses duration: -1 and ratio: adaptive, while extension and first/last-frame modes require ratio: adaptive. Generated audio, no watermark, and return_last_frame: true are the defaults. A completed task with return_last_frame: true includes last_frame_url.

1. Submit a generation task

curl https://api.hop-base.com/v1/video/generate \
  -H "Authorization: Bearer sk-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "dreamina-seedance-2-5-260628",
    "content": [
      { "type": "text", "text": "An orange cat runs across sunlit grass as the camera follows" }
    ],
    "duration": 5,
    "resolution": "720p",
    "ratio": "16:9",
    "generate_audio": false,
    "watermark": false
  }'

The response contains a task.id such as vt... for later status queries. For image-to-video, append this object to content:

{
  "type": "image_url",
  "image_url": { "url": "https://example.com/reference.jpg" },
  "role": "reference_image"
}

Other supported model parameters pass through unchanged.

2. Poll task status

# Single task; polling every 5 seconds is recommended
curl https://api.hop-base.com/v1/video/tasks/vt-your-task-id \
  -H "Authorization: Bearer sk-your-key"

# Task list; optionally append ?page=1&limit=20
curl https://api.hop-base.com/v1/video/tasks \
  -H "Authorization: Bearer sk-your-key"

status is pending, processing, completed, or failed. When complete, read the video URL from outputs. usage.completion_tokens contains the billable token count. 480p and 720p generation usually takes 2-5 minutes; 1080p and overseas-only 4K take longer.

3. Upload Seedance 2.0 reference media

curl https://api.hop-base.com/v1/sd/assets \
  -H "Authorization: Bearer sk-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "URL": "https://example.com/your-image.jpg",
    "Name": "character_front",
    "AssetType": "Image"
  }'

The returned data.Id is the asset ID. Reference it as asset://asset-id during Seedance 2.0 generation. URL must be publicly accessible. Use AssetType: Image, Video, or Audio to match the reference item, then query the asset with GET /v1/sd/assets/{asset-id} until it is ready. This upload flow applies to Seedance 2.0. Seedance 2.5 reference media must use public HTTP(S) URLs directly in content. The upload itself accepts a public URL only — no base64 or multipart — so host the file somewhere reachable first. Uploading is optional for ordinary reference images and required for any image containing a real person.

Reference media validation rules

All reference media is validated synchronously at submission. Invalid media returns an immediate 400 with a specific error code (such as invalid_image_aspect_ratio) and incurs no task cost. Pre-checking against the table below on your side eliminates most wasted calls:

Media URL must be a publicly reachable http(s) address: private / loopback / link-local addresses and URLs with credentials are rejected, with at most 3 redirects; images and audio may also be passed inline as data: base64. Role constraints: frame mode takes at most 2 images — a single image must be first_frame, two images must be exactly one first_frame and one last_frame; multimodal references must be declared explicitly as reference_image and cannot be mixed with frame mode.

Rates

Official list prices are on the pricing page. Your own rate is the one shown in the signed-in model catalog, and each request's actual charge is in your usage record.