Tencent opens early access to its HyCreator video orchestrator

HyCreator orchestrates ten-minute films from planning through quality review, with manual control available at every stage.

Producing a video several minutes long involves more than assembling separately generated clips. Characters, settings, style, narrative, and sound must remain consistent across multiple sequences. With HyCreator, Tencent Hunyuan is introducing an agentic production workspace designed to coordinate these different stages.

HyCreator is not presented as a new video generation model. Tencent describes it as an “agent harness,” meaning a layer that coordinates models, tools, and operations within a single workflow. The company has not yet specified which models handle each task or whether users will be able to select their own providers.

A project can begin with a single sentence or a more detailed brief. In automated mode, the system interprets the request, develops a plan, generates the different components, and reviews the result. Tencent’s interface displays four main stages: understand, plan, generate, and review.

This breakdown addresses a specific challenge in long-form production. A model may be able to generate a few seconds directly, but a film lasting several minutes requires numerous shots to be created and assembled. HyCreator is expected to maintain shared information across these generations, identify inconsistencies, and return to the relevant stage when something drifts.

Tencent says the system can produce films around ten minutes long without human intervention. Its public gallery includes ten demonstrations spanning narrative films, fantasy, wuxia, and visual experiments. Their listed runtimes range from 1 minute 35 seconds to 10 minutes 27 seconds, with several productions approaching the ten-minute mark.

These examples remain demonstrations selected by the team. They do not constitute a comparative evaluation and provide no indication of the system’s success rate across a wider range of briefs. Tencent has not published a testing protocol for character continuity, narrative consistency, or visual stability between shots.

Automated mode does not require users to leave the system unattended until completion. They can open the workspace and take control during production. Tencent says intervention is possible at any stage, after which the project can continue from that point instead of restarting from the original brief.

This step-by-step approach could be better suited to production than a workflow built around entirely separate generations. A failed shot, an unsuitable transition, or an inconsistent character change could theoretically be corrected without discarding elements that have already been approved. However, the exact list of editable settings and the degree of control available over each shot have not yet been documented.

HyCreator offers two production levels, Lite and Pro. The films in the gallery use one or the other, while some can be viewed in both versions. Pro is selected by default. Tencent has not yet detailed the differences in output definition, processing time, cost, or control between the two levels.

The gallery also displays the creation trace for several projects, offering a view of the steps taken by the agent and the intermediate elements produced before final assembly. This traceability could help identify where an inconsistency emerged, although Tencent has not clarified which actions users will be able to modify in the accessible version.

Several details required for professional use remain undisclosed. Tencent has provided no pricing, average production time, export formats, output definition, audio workflow, or commercial rights framework. API availability, support for third-party models, and file retention policies have not been specified either.

Tencent has also not released a technical report detailing HyCreator’s architecture. The page provides a bibliographic reference under the team’s name, but the announcement is not currently accompanied by a research paper, source code repository, or model weights. It is therefore impossible to determine how much of the output depends on the orchestration layer and how much comes from the underlying generation systems.

HyCreator is currently available through an early-access application. No general-release timeline has been announced. At this stage, its main point of interest is its approach to production: treating long-form video as a project made up of controllable stages, rather than as a single generation that must either be accepted or restarted.