Hy-Video-v1.5

What is Hy-Video-v1.5?

Hy-Video-v1.5 is Tencent’s current video-generation model in the TokenHub catalog. Its canonical API identifier is hy-video-v1.5. The model generates short video clips rather than text, images, audio, or structured data.

It supports two input workflows. A request containing only a prompt is treated as text-to-video generation. A request containing an image uses image-to-video generation, with the image serving as the visual starting point. In the latter case, the prompt can be used to describe the desired motion, although the prompt is optional according to Tencent’s documentation.

Generation is asynchronous. Instead of returning a finished video immediately, the API creates a task and returns identifiers that an application can use to check the result later. This workflow is common for media generation because rendering a video takes longer than producing a typical text response.

Supported inputs and video output

The documented output is fixed at five seconds and 720p resolution. These limits are important when evaluating the model: Hy-Video-v1.5 is designed for short clips and previews, not long-form production or configurable-duration video projects.

For text-to-video requests, Tencent documents five aspect-ratio choices:

• 16:9 for conventional landscape video
• 9:16 for vertical short-form content
• 1:1 for square social or promotional assets
• 4:3 for a traditional landscape format
• 3:4 for a traditional portrait format

Image-to-video requests work differently. They do not accept a separate aspect-ratio parameter; the output aspect ratio is derived from the supplied image. Input images can be sent as Base64 data or as publicly accessible image URLs. The documented image-size limit is 10 MB.

The supplied research does not document video-input conditioning, native audio generation, soundtrack creation, or image output. In practical terms, the supported modality combination is text and optional image input, with video as the direct model output.

Main capabilities

Hy-Video-v1.5’s most useful capability is the ability to use one model and endpoint for both prompt-based video creation and animation of a reference image. For example, a text-to-video request could describe a product reveal or a short scene transition, while an image-to-video request could animate a product photograph, illustration, or other supported still image.

The model also exposes several controls that can make generation more predictable:

• Negative prompts: descriptions of elements the output should avoid.
• Intelligent prompt rewriting: an optional feature intended to expand or refine a user’s prompt before generation.
• Content moderation controls: settings for managing moderation behavior.
• Random seed: a way to provide a seed value when reproducibility or controlled variation is useful.
• Custom watermark text: an option for adding specified watermark text to generated content.

Tencent positions the model for scene transitions, multi-character interaction, enterprise advertising, marketing material, and personal creative work. These are provider positioning claims rather than independent benchmark results. Actual quality will depend on the prompt, the source image when used, the requested composition, and the model’s handling of motion.

How the TokenHub workflow works

A typical integration sends a generation request containing the model value hy-video-v1.5, a prompt for text-to-video, and optional parameters such as an image, aspect ratio, negative prompt, seed, or watermark text. For image-to-video generation, the image can be provided in Base64 form or through a public URL.

The initial response provides a task identifier, creation timestamp, and request identifier. The application then polls the task-result endpoint until the task reaches a terminal state such as success or failure. The result is not an immediate synchronous video response, so an application should be prepared to track task status and handle delays or failed jobs.

The supplied documentation describes the API behavior and request fields but does not provide a published context window, maximum prompt-token count, maximum output-token count, throughput guarantee, or latency guarantee for this model. Those values should therefore be treated as unknown rather than inferred from other Tencent products.

Pricing and cost considerations

Tencent lists Hy-Video-v1.5 under TokenHub’s usage-based video-generation pricing. The documented rate is USD 1.60 per million tokens, with estimated consumption of 30,000 tokens per second for 720p generation. Tencent gives an equivalent reference estimate of approximately USD 0.048 per second.

Using the documented five-second duration, a simple reference calculation would be about USD 0.24 for one five-second clip at the stated estimated rate. This is an estimate, not a guaranteed per-video price: final billing depends on actual token consumption, the billing center, and the selected service configuration. The model is not documented as having a subscription tier or fixed monthly plan.

For cost planning, the model’s fixed duration can be helpful because a short clip has a relatively clear upper-level production scope. However, repeated generations, revisions, and failed or unsatisfactory attempts can raise the effective cost of a finished asset. Teams should budget for experimentation rather than assuming that one prompt will produce the final video.

Strengths and trade-offs

The strongest practical reason to consider Hy-Video-v1.5 is its focused support for short video generation in both text-to-video and image-to-video modes. A team can create a clip from a description or begin with a visual asset without adopting separate model workflows. The choice of several text-to-video aspect ratios also helps when producing content for landscape, portrait, or square placements.

The model’s main trade-off is control. Output duration is fixed at five seconds, resolution is fixed at 720p, and image-to-video aspect ratio follows the input image. These constraints simplify the documented interface but make the model less suitable when a project requires longer sequences, higher resolution, exact duration control, or independent framing of an input image.

The asynchronous API is another operational trade-off. It is appropriate for queued media jobs, but applications need task tracking, polling, timeout handling, and failure handling. Hy-Video-v1.5 should not be selected when an immediate, single-response generation flow is a strict requirement.

Editorially, the supplied model record rates its speed at 8 out of 10 and cost at 7 out of 10. These are catalog evaluations, not Tencent-published benchmark scores. They may be useful as rough comparative indicators, but they should not be interpreted as measured latency or a universal cost ranking.

Reasoning, coding, and tool support

Hy-Video-v1.5 is a media-generation model rather than a general-purpose reasoning or programming model. No reasoning score or coding score is provided, and the research does not document special reasoning or code-generation capabilities.

The API supports generation parameters, but the model record does not identify native tool use or function calling. It also does not document structured JSON output. JSON may be used by the surrounding API as a request or response format, but that is different from the model being able to produce schema-constrained structured output as a capability.

Streaming is not documented for this model. The documented interaction pattern is task creation followed by result retrieval. Fine-tuning, caching, batch processing, web search, and other general platform features are also not documented for this exact model.

Best use cases

Hy-Video-v1.5 is a reasonable choice when the required asset is a short 720p clip and the workflow benefits from either prompt-based creation or animation of an existing still image. Suitable applications include:

• Short advertising and marketing concepts
• Vertical social-media video prototypes
• Product or campaign previews
• Animating illustrations, posters, or other still creative assets
• Scene-transition experiments
• Early-stage storyboarding and creative exploration
• Short clips involving multiple characters or simple visual interactions

Image-to-video is particularly useful when a campaign already has a key visual and needs motion rather than a completely new visual composition. Text-to-video is more appropriate when the starting concept is primarily verbal and no reference image is available.

When to choose Hy-Video-v1.5

Choose Hy-Video-v1.5 when Tencent Cloud TokenHub access fits the project, five seconds is sufficient, 720p output meets the delivery requirement, and the team wants both text-to-video and image-to-video generation. It is also a good fit for workflows that can tolerate asynchronous processing and usage-based billing.

Another type of video-generation option may be more appropriate when the project requires long-form output, configurable duration, higher or selectable resolution, video-to-video transformation, native audio, or tightly controlled shot continuity. A general multimodal or language model may be preferable for prompt planning, scripting, or production coordination, but the supplied research does not identify a named Tencent sibling model that should replace Hy-Video-v1.5 for these tasks.

For image-to-video projects, choose a different option if the source image must be cropped independently of the output or if the workflow requires video input. For applications that need immediate responses, structured outputs, tool calling, or batch processing, the documented capabilities of Hy-Video-v1.5 do not establish support for those requirements.

Limitations and availability

The documented limitations are straightforward: five-second videos only, fixed 720p output, aspect-ratio selection only for text-to-video, and asynchronous task processing. The documentation also does not establish support for native audio, video input, fine-tuning, structured JSON output, or batch generation for this exact model.

Hy-Video-v1.5 is listed as current and available through Tencent Cloud TokenHub in the supplied research. Its identifier should be entered exactly as hy-video-v1.5. It should not be confused with the separately referenced older HY-Video-1.5 identifier, which is covered by a different retirement notice.