Short answer: Wan 3.0 is Alibaba's all-in-one video model, available on EmpirioLabs through one OpenAI-style endpoint. It generates clips of 2 to 30 seconds with synchronized sound, and one request can draw on up to 10 reference images, 5 reference videos, 5 audio clips, and a document or web page. Call it with model: "wan-3-0" on POST /v1/videos/generations, and pick the serving tier with model_tier: standard for the base rate or prime for significantly faster end-to-end generation.
What is new versus Wan 2.7
| Capability | Wan 2.7 | Wan 3.0 |
|---|---|---|
| Output length | 2 to 15 seconds | 2 to 30 seconds, or let the model choose |
| Reference assets | 5 combined | 10 images, 5 videos, and 5 audio clips |
| Document and web page references | Not supported | 1 file or 1 public link per request |
| Resolutions | 720p, 1080p | 480p, 720p, 1080p |
| Aspect ratio | Fixed presets | Adaptive by default, plus five presets |
| Serving tiers | One | Standard and Prime (faster end to end) |
Quickstart
Video generation is asynchronous. Submit the job, then poll it.
curl -X POST https://api.empiriolabs.ai/v1/videos/generations \
-H "Authorization: Bearer $EMPIRIOLABS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "wan-3-0",
"model_tier": "prime",
"prompt": "A hummingbird hovering over a red flower in golden hour light, slow motion",
"resolution": "1080p",
"duration": 6
}'
# -> {"job_id": "...", "status": "processing", "poll_url": "/v1/jobs/..."}
curl https://api.empiriolabs.ai/v1/jobs/JOB_ID \
-H "Authorization: Bearer $EMPIRIOLABS_API_KEY"
Three ways to drive it
- Text to video: a prompt alone, up to 20,000 characters. Output includes a generated soundtrack by default; set
audio: falsefor a silent clip at the same price. - First and last frame: attach one image to animate it, or exactly two images and the model interpolates the journey between them.
- Omni reference: attach any mix of reference images, videos, and audio, plus optionally one document or one public web link, then address them in the prompt as
Image 1,Video 1,Audio 1. The model carries subjects, motion, and voices into the new scene.
The mode parameter is optional: auto reads your attachments and picks for you.
Operational notes
- First and last frame inputs cannot be combined with reference inputs in one request. Adding audio, a document, or a web link to an image request moves it to reference mode, so an explicit
mode: "i2v"together with audio returns a validation error; leavemodeon auto and the request routes itself. - Every input image must be at least 240 pixels on each side, and at most 8,000. Smaller images are rejected before generation starts.
- With video references, input and output seconds share one 30-second budget, and billed seconds cover input plus output combined. The job result reports the exact billed duration.
duration: -1lets the model pick the clip length from your prompt and references; billing follows the actual seconds in the job result.
Pricing
Wan 3.0 is pay as you go, billed per second of video at a rate set by the resolution and the serving tier. Prime costs more per second than Standard and returns results significantly faster. Current per-second rates for all six combinations are on the model page and the pricing page.
Try it
Run it in the playground with no code, or read the API reference for the full parameter list. Wan 3.0 also powers the creative effects in the playground Effects tab. The previous generation, Wan 2.7, remains available with its video editing mode.
