Perplexity just quietly turned Computer into a video production studio. Pro and Max subscribers can now generate finished video clips, complete with audio, inside the same thread where they were already doing research, writing copy, and building files. No separate app. No switching tools mid-project.
Here is the part that makes this different from the video button you may have already noticed inside regular Perplexity search. That older feature runs on Google’s Veo 3.1 and caps out at 8 seconds per clip. What just landed inside Computer, Perplexity’s agentic workspace, is a separate and considerably more capable pipeline built on two of 2026’s most talked about video models: MiniMax H3 and ByteDance’s Seedance 2.5. If you only knew about the search bar version, you have been underestimating what this platform can now do.
What Actually Shipped
Computer already orchestrates research, writing, image generation, and file building inside a single conversation thread. It routes each task to whichever of roughly 20 specialized models fits best, and it can spin up subagents to handle pieces of a larger job on its own. Video was the obvious missing piece, and it is now filled in.
According to reporting from AlphaSignal, you can now ask Computer for a campaign clip, a product demo, or a social video, and it returns a rendered video sitting right alongside the research, copy, and images it generated for the same brief. There is no exporting your prompt to a separate video tool and reimporting the result. It happens in the thread.
Access sits where most Computer features do. Pro costs $20 a month and Max costs $200 a month, and both tiers reportedly get access to the new video capability, though Perplexity has not published exact monthly video quotas for either plan. Given how the company handles credits for its other Computer features, expect Max to come with noticeably more headroom than Pro.
MiniMax H3 vs Seedance 2.5: The Two Engines Behind This
Computer does not force you to pick a model by name. It decides which one fits your prompt. But it helps to know what each one is actually good at, because the gap between them is real.
MiniMax H3 is a 33 billion parameter model that treats text, images, video, and audio as one unified system rather than separate tools stitched together. That is why it can generate dialogue, sound effects, and ambient noise in the same pass as the visuals, instead of bolting audio on afterward. MiniMax also open sourced the model’s weights, which is a fairly unusual move for a frontier video model and one reason it is being watched closely by developers who want local deployment options.
Seedance 2.5 takes the opposite approach. It is closed and API-only, and ByteDance built it around one specific advantage: length. Where most video models cap out at four to six seconds, Seedance 2.5 generates up to 30 seconds in a single pass, and you can extend it further from there. That matters more than it sounds. A 30 second single-pass clip does not have the jarring jump cuts, lighting shifts, or wardrobe changes you get when you stitch together five separate 6 second clips to fake a longer scene.
| Feature | MiniMax H3 | Seedance 2.5 |
|---|---|---|
| Developer | MiniMax | ByteDance |
| Access | Open weights (base model) | Closed, API only |
| Clip length per pass | 4 to 15 seconds | Up to 30 seconds, extendable |
| Resolution | Native 2K via regeneration workflow | Up to 1080p, strong on continuity |
| Audio | Dialogue, SFX, and ambience in one pass, 24fps, 32kHz stereo | Synced audio, editing-focused controls |
| Best for | Short, high-resolution clips with built-in sound | Longer scenes, brand storytelling, connected shots |
If you have used either model on its own, through a dedicated platform or an API, the pitch here is that you no longer need to. Computer picks the right tool for the brief and hands you a finished clip.
How to Generate a Video Inside Perplexity Computer
You do not need to know which model does what. You just need to brief Computer the way you would brief a junior video editor. Here is the workflow.
- Open Computer from your Perplexity account. It sits in its own tab, separate from regular search, and you will need a Pro or Max subscription active to see it.
- Start with context, not just a request. Instead of “make me a video,” give it the actual job. Something closer to: “I run a cold brew coffee brand. Generate a 15 second product video showing the can on ice with condensation, then a pour shot, and write three caption options for Instagram.”
- Let it plan before it generates. For anything with multiple steps, Computer will often propose a short plan first, covering what it will research, generate, and produce. Read it. Trim it if it is doing more than you asked for, since unconstrained prompts can spin up extra subagents and burn credits on sections you never wanted.
- Attach reference material if you have it. Product photos, brand colors, an existing video clip you want matched in style. Both MiniMax H3 and Seedance 2.5 accept reference inputs, and giving them something concrete produces far more usable results than a purely text-based prompt.
- Review the output in thread. The video appears next to the rest of what Computer built for that brief: research, copy, images. Watch it before you do anything else with it.
- Ask for changes in plain language. “Make the pour shot slower,” “add ambient cafe sound,” “extend this by 10 seconds.” You are not editing a timeline yourself. You are describing what you want changed and letting the model regenerate.
- Export or hand it off. Once you are happy with the clip, download it or, if you are using Comet alongside Computer, ask the assistant to publish it directly to your social accounts.
What You Can Actually Build With This
The realistic use cases here are narrower than “anything you can imagine,” and that is actually the useful part. Where this genuinely saves time:
- Product demos. Feed it product photos and a short brief, and it can produce a demo clip alongside ad copy in the same session, instead of you managing three separate tools.
- Social campaign clips. A 15 to 30 second brand story with dialogue and sound baked in, generated from the same thread where you were already researching your audience.
- Rapid concept testing. Before you brief a real production team or freelancer, you can generate a rough version to see if the concept even works visually.
- Real estate and listings. Perplexity’s own examples include flyover-style property videos generated from a listing link, then posted with a write-up attached.
What it is not built for yet: anything requiring frame-accurate editing, multi-scene narrative with a real script, or precise brand asset placement. This is generation, not a video editor. You cannot open the timeline and nudge a clip by two frames.
What This Actually Means
I keep coming back to the same thought here. This is not really a video generation story. It is a distribution story. Perplexity did not build MiniMax H3 or Seedance 2.5. It built the orchestration layer that decides which one to use and stitched it into a workflow that already includes research and writing. That is a smart bet, because most people do not want a better video model. They want fewer tools open at once.
It also puts pressure on standalone tools like Runway, Kling, and dedicated Sora-style apps, not because Computer’s video quality necessarily beats theirs on a frame-by-frame basis, but because the friction of switching tools disappears. If you are already inside Perplexity doing research for a campaign, generating the video without leaving that thread is a real advantage, even if a specialist tool might squeeze out marginally better results on its own.
The bigger picture: MiniMax H3 and Seedance 2.5 launched within days of each other in late July 2026, and both were explicitly built to compete on production use cases rather than flashy demo reels. Perplexity folding both into one workspace, less than two months after their release, tells you how fast the video AI category is moving from “impressive toy” to “thing businesses actually route work through.”
The Honest Limitations
A few things worth knowing before you build a workflow around this.
- Quotas are not public. Perplexity has not disclosed exact monthly video generation limits for Pro versus Max inside Computer. You will find out your actual ceiling by using it.
- Quality still depends on your input. Vendor specs describe what these models can do under good conditions. Fine detail, readable on-screen text, physical object interactions, and consistent character identity across a clip remain the areas where both models can still stumble.
- No manual editing. You cannot trim, color grade, or fine-tune a generated clip inside Perplexity. You regenerate with new instructions, which is not the same thing as editing.
- Licensing and retention are unclear. Before you build recurring production work around this, it is worth confirming export rights, usage licensing, and how long Perplexity retains your generated clips.
- This is layered on external models. Perplexity is a distribution and workflow layer here, not the model creator. If MiniMax or ByteDance change pricing, availability, or terms on their end, that ripples into what you can do inside Computer.
Know someone running a small brand or content team who is still juggling four separate apps for one video? Send them this post. It might save them an afternoon.
Sources
- Perplexity Computer Adds MiniMax H3 and Seedance to Build Videos in One Thread – AlphaSignal
- Generating Videos with Perplexity – Perplexity Help Center
- Video AI: MiniMax challenges ByteDance with low price, open weights for new H3 model – South China Morning Post
- Seedance 2.5 vs MiniMax H3: Which AI Video Model Is Better in 2026 – Leadde
- MiniMax H3 vs Seedance 2.5: Which AI Video Model Should You Choose? – inSai Hilight
- Perplexity’s new Computer is another bet that users need many AI models – TechCrunch
Leave a comment