Access Wan 3.0 (Tongyi Wanxiang 3.0), the cost-first standard tier, through EvoLink's unified video API. Test text-, image-, and reference-to-video with docs and web page references first.
Pick a route, run a real generation against the Wan 3.0 API, and see the exact credits it costs before you write any code. Every run here uses the same endpoint and billing as production.
Create a video
Runs against the live API with your own balanceAdd a prompt or upload your reference material, then hit Generate — your result will appear here.
Results are kept for 24 hours — download anything you want to keep.
Live rates use the Alibaba Cloud Singapore/International price basis and bill per second of video. 480p is the base, 720p is 2x, and 1080p is 4x. Reference video seconds are added to the output duration; reference images, audio, documents, and web pages are free.
Your first test
Text-to-video, 2s at 480P — the cheapest possible request, no input video2 seconds at 480P
About 133 runs for $10
Credits are reserved when the task is created and settled against the upstream receipt when it finishes. Failed tasks are refunded in full.
Top up
At the first-test settings: 2s at 480PEnough to validate the three routes end to end
Enough for a full round of creative iteration
Enough to run a small production batch
Calculated with the same function the Playground uses, at the rate for your user group. Reference video seconds count towards billing; reference images, audio, documents, and web pages do not.
| Route | 480P | 720P(default) | 1080P | Billed seconds |
|---|---|---|---|---|
| Text-to-videowan3.0-text-to-video | $0.038/s-25% 2.55 cr/s$0.050 Official price | $0.075/s-25% 5.1 cr/s$0.100 Official price | $0.150/s-25% 10.2 cr/s$0.200 Official price | Output seconds only |
| Image-to-videowan3.0-image-to-video | $0.038/s-25% 2.55 cr/s$0.050 Official price | $0.075/s-25% 5.1 cr/s$0.100 Official price | $0.150/s-25% 10.2 cr/s$0.200 Official price | Output seconds only |
| Reference-to-videoWith image, audio, document, or web-page references | $0.038/s-25% 2.55 cr/s$0.050 Official price | $0.075/s-25% 5.1 cr/s$0.100 Official price | $0.150/s-25% 10.2 cr/s$0.200 Official price | Output seconds only |
| Reference-to-videoWith reference video | $0.038/s-25% 2.55 cr/s$0.050 Official price | $0.075/s-25% 5.1 cr/s$0.100 Official price | $0.150/s-25% 10.2 cr/s$0.200 Official price | Reference video seconds + output seconds |
Input frames and reference images are never billed — Wan 3.0 has no per-image surcharge. Turning the audio track off costs the same as leaving it on.
The minimum charge uses the 480p x 2s floor. Use the live table above for the current rate on each route.
Auto duration (duration: -1): the model picks the length, so credits are reserved at the 30s cap — about 153 credits (~$2.250) at 720P. You are only charged for the seconds actually generated; the rest is refunded when the task finishes.
Wan 3.0 is the lower-cost standard tier for 30-second multimodal video generation. Choose it for batch jobs and background pipelines; move latency-sensitive previews or user-facing generation to Wan 3.0 Prime.
Three EvoLink model IDs map onto Alibaba's wan3.0-video route - pick the ID that matches your input; every route is billed per output second.
wan3.0-text-to-videoGenerate 480p, 720p, or 1080p video from a text prompt.
wan3.0-image-to-videoAnimate a first frame, optionally with a last frame.
wan3.0-reference-videoGenerate video guided by image, video, audio, document, and web-page references.
The fields that decide how you call it, and what each one means for your integration.
wan3.0-text-to-video / wan3.0-image-to-video / wan3.0-reference-video
All three EvoLink IDs map to Alibaba's upstream wan3.0-video model. Upstream rejects mixing first/last frames with reference material, so pick the ID that matches your input.
720p
Upstream defaults to 1080p, which costs 4x the base rate. We default to 720p so a request with no quality field never lands on the most expensive tier by accident.
2-30 seconds, or -1 for auto
Auto duration lets the model choose the length. Because the length is unknown up front, credits are reserved at the 30s cap and the unused part is refunded on completion.
Up to 10 images, 5 videos, 5 audio tracks, 1 document, 1 web page
The prompt addresses them as Image 1, Video 1, Audio 1 — each modality numbered independently, in array order. A document and a web page are mutually exclusive.
On by default, switchable
The generate_audio boolean controls whether the generated video has sound. It costs the same either way. Note the field name: a top-level audio key is not part of the request schema and gets dropped, so it must be generate_audio.
Valid for about 24 hours
Result URLs point at upstream storage and are not re-hosted. Download or copy anything you need to keep before they expire.
These fields cover the main integration decisions. Open the API Reference for route-specific request and response examples.
modelstringChoose the text, image, or reference-video EvoLink model ID that matches your input.
Default: route-specific
promptstringDescribe the scene, motion, camera behavior, audio, and how numbered reference assets should be used.
Default: required
qualityenumSelect 480p, 720p, or 1080p. Resolution changes the per-second billing multiplier.
Default: 720p
durationintegerSet 2-30 output seconds, or -1 to let the model choose the duration.
Default: 5
aspect_ratioenumUse adaptive, 16:9, 4:3, 1:1, 3:4, or 9:16 for the output frame.
Default: adaptive
generate_audiobooleanTurn the generated audio track on or off without changing the video rate.
Default: true
model_paramsobjectPass first/last frames or numbered image, video, audio, document, and web-page references.
Default: route-specific
callback_urlstringReceive an HTTPS completion notification instead of relying only on task polling.
Default: optional
Use the EvoLink API for product integration and batch jobs, or call it from Codex, Claude, or Gemini for fast creative and development workflows. Both paths share the same EvoLink API key, balance, model routes, and task history.
Best for: product backends, batch jobs, automated workflows
Call EvoLink’s unified video API from your server and control the model ID, parameters, task queue, callbacks, and result storage.
Best for: creative and development tasks in Codex, Claude, and Gemini
Give the Agent your output goal, assets, and acceptance criteria. It can choose the route, assemble the request, track the task, and return the result without requiring you to hand-code every step.
The reference route is what sets Wan 3.0 apart — most of these depend on it.
This example shows the shortest runnable flow: create a task, save the returned task ID, then poll or wait for an HTTPS callback. Open the API tab for the complete parameter and response reference.
curl -X POST https://api.evolink.ai/v1/videos/generations \
-H "Authorization: Bearer $EVOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "wan3.0-text-to-video",
"prompt": "A kitten runs across a rooftop in the moonlight, city neon flickers, cinematic",
"duration": 5,
"quality": "720p",
"aspect_ratio": "16:9",
"generate_audio": true,
"callback_url": "https://your-domain.com/webhooks/video-task-completed"
}'
# Save the returned task ID, then query:
curl https://api.evolink.ai/v1/tasks/{task_id} \
-H "Authorization: Bearer $EVOLINK_API_KEY"Check every value against the parameter cards above: ranges, allowed options, asset counts, and file types must match the selected route.
Check the Authorization bearer token and confirm the available balance and task billing in the console.
Review likeness rights, brand assets, readable text, sensitive content, and input-asset compliance.
A Wan 3.0 task that reaches a final failed state is not charged. Keep the task ID, inspect the final state, and identify whether the cause is validation, review, or execution before retrying.
callback_url must be a public HTTPS URL, not localhost or a private network address. Always keep the task ID as a polling fallback.
The standard tier keeps the core 30-second multimodal workflow at a lower per-second price than Prime.
Four things that trip people up on the first integration.
Result URLs point at upstream storage with a signed expiry. Download or re-host them as part of your completion handler, not later.
With duration: -1 the reserved amount is the 30-second cap, which is 6x a 5-second request. The difference comes back on completion, but the balance check happens up front.
Each route accepts one kind of input. Sending reference videos to the image-to-video route returns 400 rather than silently ignoring them.
Generation is asynchronous and can take minutes. Poll the task endpoint or register an HTTPS callback; tasks cannot be cancelled once created.
Keep Standard for unattended batch work and choose Prime when generation time is blocking a user, creator, or reviewer. Both tiers remain asynchronous; compare the live route prices before switching.
| Decision | Wan 3.0 Standard | Wan 3.0 Prime |
|---|---|---|
| Primary advantage | Lower per-second price | Significantly faster end-to-end generation |
| Best for | Batch, background, and budget-sensitive workloads | Interactive previews, review loops, and user-facing generation |
| Core workflow | Up to 30s with text, image, or mixed references | Core Wan 3.0 capabilities aligned for the same workflow types |
| Price role | The cost-efficient default | A higher per-second price for faster turnaround |
| Choose when | Cost matters more than turnaround time | Generation time blocks the workflow |
| Speed expectation | Standard turnaround | Faster, with no fixed multiplier or latency SLA |
| Request model IDs | wan3.0-text-to-video / wan3.0-image-to-video / wan3.0-reference-video | wan3.0-prime-text-to-video / wan3.0-prime-image-to-video / wan3.0-prime-reference-video |
Images, videos, and audio are numbered independently and follow array order. Reordering an array changes what "Image 2" refers to in your prompt.
Reference video seconds are added to the output duration before the resolution multiplier. A 5s output with a 3s reference video bills as 8 seconds.
negative_prompt and prompt_extend exist on Wan 2.7 but not on Wan 3.0. Sending them returns 400 with the offending field named.
Billed per second of video, with the rate scaling by resolution. Your exact per-second rate is on the pricing page.
The same key and endpoint as every other model on EvoLink — no separate Alibaba Cloud account or workspace setup.
Credits are shared across every model. Failed tasks are refunded in full, automatically.
Poll a single task endpoint or register a callback. Reserved and settled amounts are both visible in your logs.
Supported video models share one generation endpoint and task flow. Model-specific fields still vary, so check each API reference before switching.
Compare Wan generations under one EvoLink account and API key, then choose by duration, reference material, editing features, and cost.
Wan 3.0 Prime
The speed-oriented tier for significantly faster previews, review loops, and user-facing generation.
View
Wan 2.7
The previous generation. Still the route to take if you need negative prompts, prompt enhancement, or video editing.
View
Veo 3.1
Google DeepMind's Veo 3.1: 8-second clips with native audio, available in Fast and Pro variants.
ViewWhere to go once you have decided Wan 3.0 fits.
Use the Prime tier when generation time blocks previews, reviews, or user-facing workflows.
Every Wan model available through EvoLink, side by side.
Per-second rates for every video model in one table.
One key works across every model on the platform.
Text, image, video, and audio models with live pricing.
Choose between current-generation reach and specialist editing controls.
Review availability, model IDs, pricing, and migration notes.
Wan 3.0 (Tongyi Wanxiang 3.0) is Alibaba's video generation model. EvoLink exposes three request model IDs — text-to-video, image-to-video, and reference-to-video — that map to the upstream wan3.0-video model on the standard /v1/videos/generations endpoint.
Upstream refuses to mix first/last frames with reference material in the same request. Splitting that rule into three model IDs means an invalid combination cannot be expressed in the first place, and the error you get names the right alternative.
Billing is per second and the live table above is the source of truth for your route. 480p is the base rate, 720p is 2x, and 1080p is 4x. Reference video seconds are added to the output duration.
720p. Upstream defaults to 1080p, but that tier costs four times the base rate, so we default to 720p — a request with no quality field never lands on the most expensive tier by accident.
It hands the length choice to the model. Because the output length is unknown when the task is created, credits are reserved at the 30-second cap. You are only charged for the seconds actually generated, and the difference is refunded when the task finishes.
About 24 hours. Result URLs point directly at upstream storage and are not re-hosted, so download or copy anything you want to keep as part of your completion handler.
No. Only video seconds are billed — the output duration plus any reference video seconds. Reference images, reference audio, documents, and web pages are all free.
generate_audio is a boolean that turns the generated video's soundtrack on or off, and it costs the same either way. audio_urls is an array of reference audio files on the reference-to-video route, used as material the prompt can address as Audio 1, Audio 2. Note that a top-level audio key is not accepted — the switch must be sent as generate_audio.
By number, per modality: Image 1..N, Video 1..N, Audio 1..N, each following the order of its own array. Reordering an array changes what a given number points at.
No. Wan 3.0 tasks cannot be cancelled once created. If a task fails, the reserved credits are refunded in full automatically.
No. negative_prompt and prompt_extend exist on Wan 2.7 but were dropped in Wan 3.0. Sending them returns a 400 naming the unsupported field — use Wan 2.7 if you depend on them.
Minutes rather than seconds, depending on duration and resolution. Generation is asynchronous: poll the task endpoint or register an HTTPS callback to be notified on completion.
Wan 3.0 Standard is the lower-cost choice for batch and background jobs. Wan 3.0 Prime keeps the core capabilities aligned but provides significantly faster end-to-end generation for latency-sensitive workflows. Actual completion time varies, and EvoLink does not promise a fixed speed multiplier or latency SLA.