Perplexity Computer is expanding its AI video creation capabilities with two newer video-generation models: MiniMax H3 and ByteDance’s Seedance 2.5.

The addition gives Computer users more options when generating videos, alongside the broader set of AI models that Perplexity can route between for different tasks. Perplexity announced the new video-generation options in a post on X, positioning them as another way for Computer to turn a natural-language request into a finished creative asset.

Perplexity Computer is designed to handle multi-step work rather than simply answer questions. It can plan a task, use different AI models and connected tools, and produce an output without requiring users to manually move between separate applications. Perplexity says Computer can orchestrate more than 20 models across workflows involving research, coding, design and other tasks.

With the addition of MiniMax H3 and Seedance 2.5, that workflow now has access to two video models with noticeably different capabilities.

MiniMax H3 is a multimodal video-generation model from MiniMax that can work with text, images, video and audio. The company says H3 can generate video with native stereo audio at up to 2K resolution and for up to 15 seconds. It is also designed for reference-driven generation and editing, including tasks such as transferring motion between videos and following detailed instructions.

Seedance 2.5 takes a different approach, putting more emphasis on longer clips and creative control. ByteDance says the model can generate up to 30 seconds of audio-video content in a single generation and supports multiple rounds of extension. It also adds more sophisticated reference and editing capabilities, including understanding reference video beyond simple motion transfer.

Both models generate audio together with the video, making them useful for workflows that require more than silent visual clips. That includes things such as promotional content, social videos, story-driven sequences and creative concepts where dialogue, sound effects or ambient audio need to be part of the generated result.

The additions also make sense within the direction Perplexity has been taking Computer. The service started as a way to delegate research-heavy tasks, but it has steadily expanded into a more general-purpose agent capable of producing documents, presentations, websites, images and other finished assets. Perplexity previously added video generation using Google’s Veo models, while its current subscription information lists video generation among the asset-generation capabilities available to Pro and Max users.

That means users don’t necessarily have to decide which video model to use before starting a project. Instead, Computer can become the layer between the user’s request and the underlying generation tools, depending on how the workflow is configured.

For example, a user could ask Computer to create a short product video, provide reference images and a description of the desired style, and then have the system generate the resulting asset. A more involved request could combine research, image or video generation, editing and other connected tools in the same workflow.

The arrival of MiniMax H3 and Seedance 2.5 is therefore less about Perplexity becoming another standalone AI video editor and more about video generation becoming another capability that Computer can incorporate into larger tasks.

That distinction is becoming increasingly important as AI assistants move beyond answering prompts. Perplexity’s Computer is already positioned as an agent that can take an objective and carry out the individual steps required to reach a finished result. Adding more specialized generation models gives it another set of tools to work with — and gives users more ways to turn a simple instruction into something they can actually use.