@acedatacloud/skills 2026.822.2 → 2026.905.0

package/README.md

@@ -27,6 +27,7 @@ Compatible with **30+ AI coding agents** via the [agentskills.io](https://agents
 | Skill | Description |
 |-------|-------------|
 | [flux-image](skills/flux-image/) | Generate and edit images with Flux (Black Forest Labs) |
+| [qwen-image](skills/qwen-image/) | Generate and edit images with Qwen Image 3 |
 | [seedream-image](skills/seedream-image/) | Generate and edit images with ByteDance Seedream |
 | [nano-banana-image](skills/nano-banana-image/) | Generate and edit images with Google Gemini (NanoBanana) |
 
@@ -211,6 +212,7 @@ Skills provide **knowledge** (when to use, parameters, gotchas). MCP servers pro
 | luma-video | [mcp-luma](https://pypi.org/project/mcp-luma/) | `pip install mcp-luma` | `https://luma.mcp.acedata.cloud/mcp` |
 | sora-video | [mcp-sora](https://pypi.org/project/mcp-sora/) | `pip install mcp-sora` | `https://sora.mcp.acedata.cloud/mcp` |
 | veo-video | [mcp-veo](https://pypi.org/project/mcp-veo/) | `pip install mcp-veo` | `https://veo.mcp.acedata.cloud/mcp` |
+| qwen-image | [mcp-qwen-image](https://pypi.org/project/mcp-qwen-image/) | `pip install mcp-qwen-image` | `https://qwen-image.mcp.acedata.cloud/mcp` |
 | seedream-image | [mcp-seedream](https://pypi.org/project/mcp-seedream/) | `pip install mcp-seedream` | `https://seedream.mcp.acedata.cloud/mcp` |
 | seedance-video | [mcp-seedance](https://pypi.org/project/mcp-seedance/) | `pip install mcp-seedance` | `https://seedance.mcp.acedata.cloud/mcp` |
 | happyhorse-video | [mcp-happyhorse](https://pypi.org/project/mcp-happyhorse/) | `pip install mcp-happyhorse` | `https://happyhorse.mcp.acedata.cloud/mcp` |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@acedatacloud/skills",
-  "version": "2026.822.2",
+  "version": "2026.905.0",
   "description": "Agent Skills for AceDataCloud AI services — music, image, video generation, LLM chat, web search. Compatible with Claude Code, GitHub Copilot, Gemini CLI, OpenAI Codex, and 30+ AI coding agents.",
   "keywords": [
     "agent-skills",
@@ -28,7 +28,7 @@
   },
   "repository": {
     "type": "git",
-    "url": "git+https://github.com/AceDataCloud/Skills.git"
+    "url": "https://github.com/AceDataCloud/Skills.git"
   },
   "license": "Apache-2.0",
   "author": {

package/skills/ai-chat/SKILL.md

@@ -58,7 +58,7 @@ models include:
 |--------|------------------|
 | OpenAI / reasoning | `gpt-5.6-luna`, `gpt-5.6-terra`, `gpt-5.6-sol`, `gpt-5.5`, `gpt-5.5-pro`, `gpt-5.4`, `gpt-5.4-pro`, `gpt-5.2`, `gpt-5.1`, `gpt-5`, `gpt-5-mini`, `gpt-5-nano`, `gpt-4.1`, `gpt-4.1-mini`, `gpt-4.1-nano`, `gpt-4o`, `gpt-4o-mini`, `o1`, `o3`, `o4-mini` |
 | OpenAI free-tier chat-completions | `gpt-5.5:free`, `gpt-5:free`, `gpt-4.1:free`, `gpt-4o:free`, `gpt-4o-mini:free`, `gpt-oss:free` |
-| Claude | `claude-fable-5`, `claude-opus-5`, `claude-sonnet-5`, `claude-opus-4-8`, `claude-opus-4-7`, `claude-opus-4-6`, `claude-opus-4-5-20251101`, `claude-sonnet-4-6`, `claude-sonnet-4-5-20250929`, `claude-sonnet-4-20250514`, `claude-haiku-4-5-20251001`, `claude-3-7-sonnet-20250219` |
+| Claude | `claude-fable-5-1`, `claude-fable-5`, `claude-opus-5`, `claude-sonnet-5`, `claude-opus-4-8`, `claude-opus-4-7`, `claude-opus-4-6`, `claude-opus-4-5-20251101`, `claude-sonnet-4-6`, `claude-sonnet-4-5-20250929`, `claude-sonnet-4-20250514`, `claude-haiku-4-5-20251001`, `claude-3-7-sonnet-20250219` |
 | Gemini | `gemini-3.6-flash`, `gemini-3.5-flash`, `gemini-3.1-pro`, `gemini-3.1-pro-preview`, `gemini-3.1-flash-lite-preview`, `gemini-3-pro-preview`, `gemini-2.5-flash-lite`, `gemini-2.0-flash-lite` |
 | Grok | `grok-4.5`, `grok-4`, `grok-4-0709`, `grok-3`, `grok-3-fast` |
 | DeepSeek | `deepseek-v4-pro`, `deepseek-v4-flash`, `deepseek-r1`, `deepseek-r1-0528`, `deepseek-v3`, `deepseek-v3-250324`, `deepseek-v3.2-exp` |

package/skills/fish-audio/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: fish-audio
-description: Generate AI text-to-speech audio with Fish Audio and browse public reference voices via AceDataCloud API. Use when creating voiceover/narration audio (TTS), synthesizing multilingual speech, or selecting a Fish reference voice from the model catalog.
+description: Generate AI text-to-speech audio, use saved voices, or create a one-shot voice clone from an HTTPS reference audio URL and exact transcript via AceDataCloud API.
 license: Apache-2.0
 metadata:
   author: acedatacloud
@@ -66,7 +66,28 @@ Headers:
 }
 ```
 
-### 3. Async TTS
+### 3. One-shot voice cloning
+
+Use a temporary reference voice without creating a persistent model:
+
+```json
+POST /fish/tts
+Headers:
+  model: s2-pro
+
+{
+  "text": "New speech in the referenced voice.",
+  "format": "mp3",
+  "references": [{
+    "audio": "https://cdn.acedata.cloud/reference.mp3",
+    "text": "The exact words spoken in the reference audio."
+  }]
+}
+```
+
+`audio` must be a public HTTPS MP3/WAV URL and `text` must be the exact transcript. Use one reference lasting 10–270 seconds. Do not combine `references` with `reference_id`; use `reference_id` when the same saved/public voice will be reused. Raw bytes, Base64, data URIs, and MessagePack are not accepted by the AceDataCloud endpoint.
+
+### 4. Async TTS
 
 ```json
 POST /fish/tts
@@ -105,7 +126,7 @@ Headers:
 | `max_new_tokens` | integer | Maximum generated tokens |
 | `normalize` | boolean | Normalize generated audio |
 | `prosody` | object | Prosody tuning |
-| `references` | array | Additional reference objects |
+| `references` | array | One `{audio, text}` object for a one-shot voice clone; mutually exclusive with `reference_id` |
 | `callback_url` | string | Async callback URL |
 | `async` | boolean | Run asynchronously and poll `/fish/tasks` |
 
@@ -114,4 +135,6 @@ Headers:
 - The documented TTS endpoint is `POST /fish/tts` — not `/fish/audios`.
 - Choose the Fish engine with the **`model` request header**, not a JSON `model` field.
 - Use `reference_id` from `GET /fish/model` — not `voice_id`.
+- Use `references` for a one-shot clone that is not saved as a model.
+- Billing is based on the target text UTF-8 byte count; the reference audio does not add a separate clone fee.
 - Synchronous requests return `audio_url` directly; async jobs should be polled via `/fish/tasks`.

package/skills/maestro-video/SKILL.md

@@ -34,7 +34,6 @@ curl -X POST https://api.acedata.cloud/maestro/videos \
     "prompt": "Create a 30-second beginner-friendly video explaining vector databases. End with one memorable takeaway.",
     "aspect": "16:9",
     "duration": 30,
-    "quality": "standard",
     "scenario": "narrated",
     "langs": ["en"]
   }'
@@ -72,29 +71,23 @@ Polling and history queries are free; the video task is settled after production
 | `file_urls` | string[] | - | Public image, video, or audio references, up to the limits enforced by the API |
 | `langs` | string[] | `["zh-cn"]` | Output language codes; each creates a localized rendered variant |
 | `aspect` | string | `9:16` | `9:16`, `16:9`, or `1:1` |
-| `duration` | integer | `30` | Target duration from 5 to 300 seconds; the selected SKU sets the maximum |
-| `quality` | string | `standard` | `lite`, `standard`, or `pro` |
-| `scenario` | string | `auto` | `auto`, `narrated`, `captions`, `avatar`, or `drama`; availability depends on SKU |
+| `duration` | integer | `30` | Target duration from 5 to 300 seconds |
+| `scenario` | string | `auto` | `auto`, `narrated`, `captions`, `avatar`, or `drama` |
 | `style` | string | `auto` | Preset or freeform visual direction |
 | `voice` | string | `auto` | Cross-lingual voice preset or a 32-hex-character Fish reference ID |
 | `callback_url` | string | - | Optional webhook called on success or failure |
 
-### Production SKUs
+### Production contract
 
-| SKU | Price | Duration | Output | Languages | Scenarios and actions |
-|---|---:|---:|---|---:|---|
-| `lite` | 0.20 Credits/second | 5–30s | 720p/24fps | 1 | auto/narrated/captions; generate/edit |
-| `standard` | 0.60 Credits/second | 5–120s | 1080p/30fps | 2 | adds avatar and remix |
-| `pro` | 1.20 Credits/second | 5–300s | 1080p/30fps | 4 | adds drama and extend |
+Every request uses the complete production capability: all actions and scenarios, 5–300 seconds, up to 4 languages, and 1080p/30fps output. The base price is 0.60 Credits per delivered second. Avatar uses a 1.15× multiplier, drama uses 1.35×, and each additional delivered language adds 6 Credits. Failed tasks and polling are free.
 
-Successful tasks are billed by delivered integer-second duration, with no 30-second minimum. Avatar uses a 1.15× multiplier, drama uses 1.35×, and each additional delivered language adds 6 Credits. Failed tasks and polling are free.
 
 ### Scenarios
 
 - `auto`: let the director choose from the brief.
 - `narrated`: multi-scene explainer, documentary, brand, history, or product video with voiceover.
 - `captions`: add kinetic captions to a source video supplied in `file_urls`.
-- `avatar`: talking-head or digital-human video; provide a portrait in `file_urls` (Standard or Pro).
+- `avatar`: talking-head or digital-human video; provide a portrait in `file_urls`.
 - `drama`: acted short drama with characters and dialogue (Pro only).
 
 ### Styles

package/skills/qwen-image/SKILL.md

@@ -0,0 +1,59 @@
+---
+name: qwen-image
+description: Generate and edit images with Qwen Image 3 via Ace Data Cloud. Use for text-to-image, 1-3 reference-image editing, batch output, precise text rendering, or asynchronous image tasks.
+license: Apache-2.0
+metadata:
+  author: acedatacloud
+  version: "1.0"
+compatibility: Requires ACEDATACLOUD_API_TOKEN. Optionally pair with mcp-qwen-image.
+---
+
+# Qwen Image 3
+
+Use `POST https://api.acedata.cloud/qwen-image/images` for generation and editing.
+
+## Models
+
+| Model | Best for |
+|---|---|
+| `qwen-image-3.0` | Value, throughput, batch production |
+| `qwen-image-3.0-pro` | Complex layouts, small text, fine detail |
+
+## Generate
+
+```bash
+curl -X POST https://api.acedata.cloud/qwen-image/images \
+  -H "Authorization: Bearer $ACEDATACLOUD_API_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{"model":"qwen-image-3.0","prompt":"A clean bilingual product poster","size":"1024*1024","n":1}'
+```
+
+## Edit with references
+
+```json
+{
+  "model": "qwen-image-3.0-pro",
+  "prompt": "Keep the subject and change the scene to a cinematic city at night",
+  "image_urls": ["https://cdn.acedata.cloud/r9vsv9.png"],
+  "size": "2048*2048",
+  "n": 1,
+  "prompt_extend": true,
+  "enable_thinking": true,
+  "watermark": false
+}
+```
+
+Use 1–3 reference images. `prompt_extend_mode=agent` is text-to-image only. `n` supports 1–6. Output size uses `WIDTH*HEIGHT`, with pixel area between 512×512 and 2048×2048 and aspect ratio between 1:8 and 8:1.
+
+## Async tasks
+
+Set `async: true` or provide `callback_url`. Poll the returned task with:
+
+```json
+POST /qwen-image/tasks
+{"action":"retrieve","id":"TASK_ID"}
+```
+
+Task retrieval is free. Keep polling every 15 seconds until the response is terminal.
+
+> **MCP:** `pip install mcp-qwen-image` | Hosted: `https://qwen-image.mcp.acedata.cloud/mcp`

package/skills/seedream-image/SKILL.md

@@ -1,115 +1,123 @@
 ---
 name: seedream-image
-description: Generate and edit AI images with Seedream (ByteDance) via AceDataCloud API. Use when creating images from text prompts, editing existing images, or working with high-resolution outputs. Supports Seedream 4.0, 4.5, and 5.0 models.
+description: Generate, edit, stream, or decompose images with Seedream 5.0 via AceDataCloud. Use for text-to-image, reference-image editing, related image sets, transparent-background edits, or editable layer extraction.
 license: Apache-2.0
 metadata:
   author: acedatacloud
-  version: "1.0"
-compatibility: Requires ACEDATACLOUD_API_TOKEN in .env file (see _shared/authentication.md). Optionally pair with mcp-seedream for tool-use.
+  version: "1.1"
+compatibility: Requires ACEDATACLOUD_API_TOKEN (see ../_shared/authentication.md). Optionally pair with mcp-seedream-pro or seedream-cli.
 ---
 
-# Seedream Image Generation
+# Seedream Image
 
-Generate and edit AI images through AceDataCloud's Seedream (ByteDance) API.
+Use `POST https://api.acedata.cloud/seedream/images`. Authenticate with `Authorization: Bearer $ACEDATACLOUD_API_TOKEN` and JSON request bodies.
 
-> **Setup:** See [authentication](../_shared/authentication.md) for token setup.
+## Pick the model from the requested capability
 
-## Quick Start
+| Capability | Seedream 5.0 Pro | Seedream 5.0 Lite |
+|---|---|---|
+| Model ID | `doubao-seedream-5-0-pro-260628` | `doubao-seedream-5-0-260128` (alias: `doubao-seedream-5-0-lite-260128`) |
+| Generate/edit one image | Yes | Yes |
+| Reference images | Up to 10 | Up to 14 |
+| Related image set | No | Yes; input + output ≤ 15 |
+| Streaming / web search | No | Yes |
+| Layer decomposition / transparent background | Yes | No |
+| Prompt optimization | `standard`, `fast` | `standard` |
+| Preset sizes | `1K`, `1.5K`, `2K` | `2K`, `3K`, `4K` |
+
+Seedream 4.5 and 4.0 remain available for compatibility. Do not send a parameter to a model that does not support it.
+
+## Generate or edit
 
 ```bash
-curl -X POST https://api.acedata.cloud/seedream/images \
+curl https://api.acedata.cloud/seedream/images \
   -H "Authorization: Bearer $ACEDATACLOUD_API_TOKEN" \
   -H "Content-Type: application/json" \
-  -d '{"prompt": "a cyberpunk cat wearing VR goggles in a neon city", "model": "doubao-seedream-5-0-260128"}'
+  -d '{
+    "model": "doubao-seedream-5-0-260128",
+    "prompt": "a four-panel storyboard of a courier crossing a rainy neon city",
+    "size": "2K",
+    "sequential_image_generation": "auto",
+    "sequential_image_generation_options": {"max_images": 4},
+    "watermark": false
+  }'
 ```
 
-> **Async:** See [async task polling](../_shared/async-tasks.md). Poll via `POST /seedream/tasks` with `{"id": "..."}`.
-## Models
-
-| Model | Version | Best For |
-|-------|---------|----------|
-| `doubao-seedream-5-0-260128` | Seedream 5.0 | Latest, highest quality (default) |
-| `doubao-seedream-5-0-pro-260628` | Seedream 5.0 Pro | Highest-quality Seedream 5.0 model |
-| `doubao-seedream-4-5-251128` | Seedream 4.5 | High quality, balanced |
-| `doubao-seedream-4-0-250828` | Seedream 4.0 | Reliable generation |
-
-## Workflows
+For editing, add `image` as one URL/Base64 string or an array. `response_format` is `url` or `b64_json`; 5.0 Pro/Lite also support `output_format` as `jpeg` or `png`. Explicit dimensions use `WIDTHxHEIGHT` and must satisfy the selected model's pixel and aspect-ratio limits.
 
-### 1. Text-to-Image
+Use Lite web search only when current information matters:
 
 ```json
-POST /seedream/images
-{
-  "prompt": "a serene Japanese garden with cherry blossoms and a red bridge",
-  "model": "doubao-seedream-5-0-260128",
-  "size": "1K"
-}
+{"tools": [{"type": "web_search"}]}
 ```
 
-### 2. Image Editing (Image-to-Image)
+## Transparent-background Pro edit
 
-Edit an existing image by providing the source image URL(s) and a descriptive prompt. Seedream 5.0 Pro, 5.0 Lite, 4.5, and 4.0 all support image input.
+Use one transparent PNG input and PNG output:
 
 ```json
-POST /seedream/images
 {
-  "prompt": "change the sky to a golden sunset",
-  "model": "doubao-seedream-5-0-260128",
-  "image": ["https://example.com/photo.jpg"]
+  "model": "doubao-seedream-5-0-pro-260628",
+  "prompt": "replace the parrot with a peacock",
+  "image": "https://example.com/layer.png",
+  "background": "transparent",
+  "output_format": "png",
+  "size": "1.5K"
 }
 ```
 
-### 3. Async Generation with Task Polling
+Do not combine `background` with layer decomposition. `transparent` with JPEG is invalid.
 
-Pass a `callback_url` to receive results asynchronously via webhook, or poll `/seedream/tasks` for the result:
+## Decompose an image into editable layers
 
-```json
-POST /seedream/images
-{
-  "prompt": "an epic fantasy landscape",
-  "model": "doubao-seedream-5-0-260128",
-  "callback_url": "https://api.acedata.cloud/health"
-}
+`layer_decomposition` is Pro-only and requires exactly one PNG/JPEG. Omit `prompt` for automatic decomposition, describe desired elements in natural language, or use normalized `<bbox>` coordinates.
+
+```bash
+curl https://api.acedata.cloud/seedream/images \
+  -H "Authorization: Bearer $ACEDATACLOUD_API_TOKEN" \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "doubao-seedream-5-0-pro-260628",
+    "image": "https://example.com/poster.png",
+    "layer_decomposition": true,
+    "size": "auto",
+    "watermark": false
+  }'
 ```
 
-Poll the returned `task_id`:
+The response contains one base image (`z_index: 0`) and up to 16 transparent PNG layers. Each layer can include `name`, `description`, and `bounding_box.absolute`/`normalized`. To recompose, scale each layer to its bounding-box width and height, place it at left/top, and draw in ascending `z_index`. Any layer failure fails the whole decomposition.
+
+## Async tasks
+
+Generation can take time. Prefer `"async": true`; the response returns `task_id`. Poll using:
 
 ```json
 POST /seedream/tasks
-{"id": "<task_id>"}
+{"action": "retrieve", "id": "<task_id>"}
 ```
 
-## Parameters
+Follow [async task polling](../_shared/async-tasks.md). Use `callback_url` only when a real public webhook exists. Do not use a health endpoint as a fake callback.
+
+## Streaming
 
-### Generation
+Lite/4.x support streaming. Send `"stream": true` with `Accept: application/x-ndjson`; read one normalized JSON event per line until `image_generation.completed`. Do not combine streaming with `async` or `callback_url`.
 
-| Parameter | Values | Description |
-|-----------|--------|-------------|
-| `model` | see Models table | Model to use (required) |
-| `prompt` | string | Image description (required) |
-| `size` | `"1K"`, `"2K"`, `"3K"`, `"4K"`, `"adaptive"` | Output resolution; available presets depend on the selected model |
-| `sequential_image_generation` | `"auto"`, `"disabled"` | Generate related images in sequence (5.0, 4.5, 4.0 only) |
-| `stream` | boolean | Stream images as they're generated (5.0, 4.5, 4.0 only) |
-| `watermark` | boolean | Add AI-generated watermark (default: true) |
-| `output_format` | `"jpeg"`, `"png"` | Output file format (Seedream 5.0 only; default: jpeg) |
-| `response_format` | `"url"`, `"b64_json"` | Response format (default: url) |
-| `tools` | array | Enable tools, e.g. `[{"type": "web_search"}]` (Seedream 5.0 only) |
-| `callback_url` | string | Webhook URL for async delivery; returns `task_id` immediately |
+- `image_generation.partial_succeeded`: one generated image
+- `image_generation.partial_failed`: one failed item; other Lite images may still succeed
+- `image_generation.completed`: final usage and the only billing completion
 
-### Editing
+## Agent tools
 
-| Parameter | Required | Description |
-|-----------|----------|-------------|
-| `image` | Yes (for editing) | Array of image URLs or base64 strings (max 10MB each) |
-| `prompt` | Yes | Describe the desired edit |
+- MCP: `pip install mcp-seedream-pro`; use `seedream_generate_image`, `seedream_edit_image`, `seedream_decompose_image`, then poll with `seedream_get_task`.
+- CLI: `pip install seedream-cli`; use `seedream generate`, `seedream edit`, `seedream decompose`, or `seedream generate --stream --json`.
+- Hosted MCP: `https://seedream.mcp.acedata.cloud/mcp`.
 
-## Gotchas
+MCP image tools are asynchronous, so use REST or CLI for real-time streaming.
 
-- Model names now use the `doubao-*` naming convention (e.g. `doubao-seedream-5-0-260128`)
-- Image editing uses the same `/seedream/images` endpoint with the `image` array parameter (no separate edit endpoint)
-- `size` replaces separate `width`/`height` params; use `"1K"` for 1024×1024, `"2K"` for 2048×2048, etc.
-- 5.0 Pro supports `1K`/`2K`; 5.0 Lite supports `2K`/`3K`/`4K`; 4.5 supports `2K`/`4K`; 4.0 supports `1K`/`2K`/`4K`; `adaptive` selects a size from the reference image
-- `stream` and `sequential_image_generation` are only available for Seedream 5.0, 4.5, and 4.0
-- Pass `callback_url` to get a `task_id` immediately and avoid blocking; poll `/seedream/tasks` for the result — use `"https://api.acedata.cloud/health"` as a placeholder to force async mode without a real webhook
+## Result and billing safeguards
 
-> **MCP:** `pip install mcp-seedream` | Hosted: `https://seedream.mcp.acedata.cloud/mcp` | See [all MCP servers](../_shared/mcp-servers.md)
+- URL results expire; persist files promptly.
+- Lite `data[]` may mix successful images and per-item `error` objects. Count only successful images.
+- Pro layer items are billed individually by actual output size; the base and every successful layer count separately.
+- Preserve `z_index`, bounding boxes, item errors, `tools`, and `usage`; do not flatten the response to a URL list.
+- Never estimate final billed cost from requested size alone. Use the returned `cost`/usage and the live pricing page: https://platform.acedata.cloud/services/seedream?tab=pricing.

package/skills/wan-video/SKILL.md

@@ -32,6 +32,7 @@ curl -X POST https://api.acedata.cloud/wan/videos \
 | `wan2.6-i2v` | Image-to-Video | Animating a still image into video |
 | `wan2.6-r2v` | Reference Video-to-Video | Character extraction and transfer from reference video |
 | `wan2.6-i2v-flash` | Image-to-Video (Fast) | Quick image-to-video generation |
+| `wan3.0-video` | All-in-One | Text, frames, reference media, files, and public links |
 
 ## Workflows
 
@@ -150,3 +151,24 @@ POST /wan/videos
 - `shot_type: "multi"` produces multi-cut edits rather than a single continuous shot
 
 > **MCP:** `pip install mcp-wan` | Hosted: `https://wan.mcp.acedata.cloud/mcp` | See [all MCP servers](../_shared/mcp-servers.md)
+
+
+## Wan 3 All-in-One
+
+`wan3.0-video` infers the workflow from `media`. Supported types are `first_frame`, `last_frame`, `reference_image` (up to 10), `reference_video` (up to 5, 15 seconds total), `reference_audio` (up to 5, 15 seconds total), `file`, and `link`. Frame mode cannot be mixed with reference/file/link mode.
+
+```json
+POST /wan/videos
+{
+  "model": "wan3.0-video",
+  "prompt": "Use image 1 to create a cinematic product video",
+  "media": [{"type":"reference_image","url":"https://cdn.acedata.cloud/r9vsv9.png"}],
+  "resolution": "720P",
+  "ratio": "16:9",
+  "duration": 5,
+  "audio": true,
+  "async": true
+}
+```
+
+Wan 3 supports 2–30 seconds or `-1` automatic duration. Poll with the existing `/wan/tasks` endpoint.