aliyun-kling-video

这段内容是审核时保存的快照。外部来源才是完整且最新的版本。

---
name: aliyun-kling-video
description: Use when generating videos with Kling v3 models on DashScope (kling/kling-v3-video-generation, kling/kling-v3-omni-video-generation). Use when implementing text-to-video, image-to-video, reference-to-video, smart storyboard, or video editing via the video-synthesis async API.
---

# Kling V3 Video Generation

## Validation

```bash
mkdir -p output/aliyun-kling-video
python -m py_compile skills/ai/video/aliyun-kling-video/scripts/generate_kling_video.py && echo "py_compile_ok" > output/aliyun-kling-video/validate.txt
```

Pass criteria: command exits 0 and `output/aliyun-kling-video/validate.txt` is generated.

## Output And Evidence

- Save task IDs, polling responses, and final video URLs to `output/aliyun-kling-video/`.
- Keep at least one end-to-end run log for troubleshooting.

## Prerequisites

- Install dependencies (recommended in a venv):

```bash
python3 -m venv .venv
. .venv/bin/activate
python -m pip install requests
```
- Set `DASHSCOPE_API_KEY` in your environment (must be Beijing region API Key).
- Enable Kling in [百炼控制台](https://bailian.console.aliyun.com/cn-beijing/?tab=model#/model-market/all) — search "kling" and activate.

## Critical model names

- `kling/kling-v3-video-generation` — standard model: t2v, i2v (first frame, first+last frame)
- `kling/kling-v3-omni-video-generation` — omni model: adds reference-to-video, video editing, multi-subject references

## Capabilities

| Capability | Model | Required media |
|---|---|---|
| Text-to-video | both | none |
| Smart storyboard (multi-shot) | both | none (use `multi_prompt`) |
| Image-to-video (first frame) | both | `first_frame` |
| Image-to-video (first+last frame) | both | `first_frame` + `last_frame` |
| Reference-to-video | omni only | `refer` and/or `feature` |
| Video editing | omni only | `base` + optional `refer` |

## API endpoint (async only)

```
POST https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis
```

Required headers:
- `Authorization: Bearer $DASHSCOPE_API_KEY`
- `Content-Type: application/json`
- `X-DashScope-Async: enable`

**Region**: Beijing only. No Singapore endpoint.

## Normalized interface

### Request (input)
- `prompt` (string, conditional) — up to 2500 characters. Required for `shot_type=intelligence`. For omni reference-to-video, use `<<<element_1>>>`, `<<<image_1>>>`, `<<<video_1>>>` to reference media.
- `negative_prompt` (string, optional) — content to exclude
- `media` (array, optional) — media objects with `type` and `url`:
  - Standard model types: `first_frame`, `last_frame`
  - Omni model types: `first_frame`, `last_frame`, `refer`, `base`, `feature`
- `multi_shot` (boolean, optional) — enable multi-shot generation (default: false)
- `shot_type` (string, conditional) — `intelligence` (AI auto-split) or `customize` (manual). Required when `multi_shot=true`.
- `multi_prompt` (array, optional) — per-shot prompts when `shot_type=customize`
- `element_list` (array, optional) — multi-subject element images (omni model only)
- `keep_original_sound` (string, optional) — `no` (default) or `yes`, for videos (omni model only)

### Request (parameters)
- `mode` (string, optional) — `pro` (default, 1080P) or `std` (720P)
- `aspect_ratio` (string, conditional) — `16:9` (default), `9:16`, `1:1`. Required for t2v and reference-to-video.
- `duration` (integer, optional) — video length [3, 15] seconds (default: 5). When using reference video, [3, 10].
- `audio` (boolean, optional) — generate audio (default: false). Affects pricing.
- `watermark` (boolean, optional) — add "可灵 AI" watermark (default: false)

### Media input limits

**Images** (first_frame, last_frame, refer):
- Formats: JPEG, JPG, PNG (no transparency)
- Resolution: [300, 8000] pixels per side
- Max size: 10MB

**Videos** (base, feature):
- Formats: mp4, mov
- Duration: 3-10s
- Resolution: [720, 2160] pixels per side
- Frame rate: 24-60 fps
- Max size: 200MB

### Media combination rules

**kling/