# Tencent Cloud WAND

> Source: https://aiwiki.ai/wiki/tencent_wand
> Updated: 2026-11-04
> Categories: AI Tools & Products, Image Generation, Multimodal AI, Video Generation
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Tencent Cloud WAND." aiwiki.ai, 4 Nov 2026. https://aiwiki.ai/wiki/tencent_wand
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**Tencent Cloud WAND** is [Tencent](https://aiwiki.ai/wiki/tencent) Cloud's brand for AI capabilities used in audio and video services. Its WAND-Vega family includes image and video generation, multimodal understanding and audio processing. The generation services are available through TokenHub APIs with named model tiers and asynchronous tasks.[1][2][3]

## Product scope

WAND combines specialist and general-purpose models with processing software. Its service interfaces include Media Processing Service (MPS), Video on Demand (VOD), Cloud Streaming Services (CSS) and TokenHub.[1]

| Family | Function described by Tencent Cloud |
|---|---|
| WAND-Vega-Video | Video generation from text, frames or reference media |
| WAND-Vega-Image | Image generation from text and reference images |
| WAND-Vega-VL | [Multimodal understanding](https://aiwiki.ai/wiki/multimodal_model), including video summaries, product classification and embodied-AI data labeling |
| WAND-Vega-Audio | [Speech recognition](https://aiwiki.ai/wiki/speech_recognition), dubbing, subtitle translation, voice separation and audio restoration |
| WAND-Codec+ | Semantic analysis to guide encoding, bitrate allocation and compression |
| WAND-Enhance+ | [Diffusion Transformer](https://aiwiki.ai/wiki/diffusion_transformer) image reconstruction, super-resolution, frame interpolation and artifact reduction |
| WAND-EraseVibe+ | Generative removal and reconstruction of obscured image regions |

These functions can form pipelines. Tencent's sports-broadcast example combines visual analysis, spoken commentary, highlights, reframing, subtitles and encoding. AIGC Studio also combines model integrations with prompt processing, output repair and retries. Such workflows involve multiple components.[1]

## Video generation API

The international guide lists **WAND-Vega-Video-1.0-Lite**, API identifier `wand-vega-video-1.0-lite`, for [text-to-video](https://aiwiki.ai/wiki/text_to_video), first/last-frame and reference-based generation.[2]

| Setting | Documented value |
|---|---|
| Duration | Integer from 4 to 15 seconds; default 5 |
| Resolution | `768P`, `1080P`, `2K`, `4K` |
| Reference media | Image, video and audio URLs |
| Prompt | At least one text element is required |

Requests use bearer authentication and require TokenHub activation and the corresponding model's postpaid billing. Under `https://tokenhub-intl.tencentcloudmaas.com`, clients submit `POST /v1/wand/vega-videos/tasks` and poll `GET /v1/wand/vega-videos/tasks/{id}`. Submission returns HTTP 201, an `id` and `queued`. Subsequent states include `running`, `succeeded` and `failed`; successful results appear in `content.video_url`.[2]

First/last-frame inputs allow at most one image per role. For reference generation, the guide recommends at most nine images, three videos and three audio clips, with twelve assets total. Base64 inline media is unsupported. Generated video URLs expire after twelve hours. Each submission creates a new task: the endpoint has no request-level idempotency, so retry handling must prevent duplicate generation and charges. Queue timeouts can return `TaskTimeout`.[2]

## Image generation API

WAND-Vega-Image supports [text-to-image and reference-image generation](https://aiwiki.ai/wiki/ai_image_generation). Its documented tiers use the identifiers `wand-vega-image-lite`, `wand-vega-image-flash` and `wand-vega-image-pro`.[3]

| Setting | Documented value |
|---|---|
| Prompt limit | 2,000 characters |
| Resolution tiers | 1K, 2K and 4K, with preset dimensions by aspect ratio |
| Reference-image count | Up to three for Lite; up to six for Flash and Pro |
| Reference-image format | PNG or JPEG; maximum 10 MB |

Clients submit `POST /v1/wand/vega-images/generations` on the same international host and receive a `task_id`. They poll `GET /v1/wand/vega-images/tasks/{task_id}`; the guide recommends intervals of three to five seconds. Successful image tasks use `completed`, and image URLs appear in `data[].url`. Other states include `queued`, `in_progress`, `failed`, `cancelled` and `incomplete`.[3]

These image-task fields differ from the video endpoint's `id`, `succeeded` and `content.video_url`. Image download URLs are temporary signed addresses that should be saved promptly; the image guide does not specify the video's twelve-hour lifetime for them.[2][3]

## References

1. Tencent Cloud. [WAND product overview](https://cloud.tencent.com/product/wand). Chinese-language product and workflow descriptions. Accessed November 4, 2026.
2. Tencent Cloud. [WAND Video Generation API Guide](https://intl.cloud.tencent.com/document/product/1300/84930). Updated October 16, 2026. Accessed November 4, 2026.
3. Tencent Cloud. [WAND Image Generation API Guide](https://intl.cloud.tencent.com/document/product/1300/83859). Updated September 16, 2026. Accessed November 4, 2026.