An AI News Network Went Live on X With No Human on Camera. It Cost About $50,000 for the Day.

On August 12, Mirage put four AI-generated news anchors on X and let them run for 24 hours straight. No human face, no human voice, nobody on set. Reported cost for the day: roughly $50,000. The stream pulled about 50,000 viewers and the launch post hit 824,000 impressions. Mirage is the AI video company formerly known as Captions, and it raised $75 million five months earlier, so this wasn't a scrappy weekend project. It's worth walking through what it actually cost, why, and what part of it (if any) a solo operator can realistically touch.

What actually happened

Mirage streamed its "news network" live on X through the @TryMirage account, with four synthetic anchors reading real news stories in an intentionally retro, '80s-broadcast style, complete with names like Tom Callahan and Marcus Sterling. The anchors were built by mixing Mirage's own Avatar X model with ChatGPT's image generator and Google's Gemini text-to-speech. CEO Gaurav Misra has said he built and ran the project largely by himself, with about $30,000 of that reported $50,000 going toward Claude Code work, meaning the bulk of the spend was engineering and orchestration, not the on-screen media itself. Misra has also been explicit that the goal wasn't to fool anyone: making sure viewers knew the anchors were synthetic was, in his words, the most important part of the project.

Mirage isn't new to this space and isn't a startup running on fumes. It was founded in 2021 by Gaurav Misra and Dwight Churchill, both former Snap engineers, and it rebranded from Captions to Mirage earlier this year to position itself as a broader AI lab rather than just a mobile editing app. It closed $75 million in growth financing from General Catalyst's Customer Value Fund in March 2026, putting total funding at roughly $175 million. It's also not the first company to try AI news anchors: Channel 1 announced a similar concept back in 2023 and never got to sustained live broadcasting at scale, Kuwait's state news agency has had its AI presenter "Fedha" since 2023, and China's Xinhua has run a virtual anchor since 2018. What's different here is that Mirage actually kept a live feed running continuously for a full day, on a platform with real reach, rather than shipping a demo reel.

The $50,000 is an infrastructure bill, not a content budget

Here's where the headline number gets misread. $50,000 sounds like a lot to spend on "a day of AI video," and the instinct is to compare it to what you'd pay a video editor or a small production crew. That's the wrong comparison. Almost all of that cost came from running generation continuously and in real time for 24 hours: constant avatar rendering, constant text-to-speech, a pipeline that had to keep pulling in stories and pushing out video without stopping, plus someone (Misra, by his own account) actively managing it live. Real-time, always-on synthetic media is a fundamentally different and much more expensive problem than producing a finished clip you render once and publish.

That distinction matters more than the dollar figure. If Mirage had pre-rendered twenty-four one-hour segments and scheduled them to play back sequentially, the compute cost would have dropped enormously, because rendering is a discrete, bounded job instead of an open-ended live stream you have to keep feeding. The $50,000 buys you the "live" part, and live is the expensive part. Nothing about this number tells you what it costs to make one AI-anchored video and post it.

What the actual solo-operator version of this costs

This is the part worth sitting with. The tools that make a single AI-narrated video are not new, not exotic, and not remotely close to $50,000. HeyGen's Creator plan runs $29 a month and includes 600 credits, with its higher-quality Avatar IV output costing 20 credits per minute, so a solo operator gets a real chunk of avatar video a month for less than a streaming subscription. ElevenLabs' Creator tier is $22 a month and unlocks professional voice cloning. Put those together and you have a stack that can produce a two-minute, scripted, AI-narrated video for well under $60 a month in tooling, published whenever you want, with no live infrastructure to babysit.

I've used one of these tools exactly once, HeyGen, for a 45-second product demo. It took about twenty minutes start to finish and looked convincing at that length. I haven't touched anything designed to run continuously, and after reading how Mirage's day actually went, I'm not in a hurry to. The gap between "render a two-minute clip" and "keep a synthetic anchor talking live for 24 hours without a glitch" is the entire story here, and it's a gap most solo operators will never need to cross, because the async version is where the actual content value is anyway.

What this demonstrates that's genuinely useful

The part of Mirage's stunt that's worth paying attention to isn't "AI news anchors are here." It's confirmation of where the real-time frontier currently sits: continuous, live synthetic video generation is still hard and expensive enough that a company with $175 million in funding needed its CEO personally running the pipeline and burning tens of thousands of dollars to keep one day of it stable. Meanwhile the async version of the same underlying models, avatar generation, voice cloning, lip sync, has quietly gotten cheap and reliable enough that anyone can rent it by the month. If you want a synthetic presenter for a niche newsletter recap, a weekly product update, or a daily "here's what happened" clip in your industry, that's a solved, affordable problem today. A live 24-hour channel is not, and won't be for most builders for a while.

What I'd actually do

If this made you want to build something, don't build a live channel. Pick a narrow, recurring content format, a two-to-five-minute weekly or daily recap in a niche you actually understand, script it yourself so it says something real, and render it with HeyGen or a comparable tool plus a cloned or licensed voice. Publish it as a normal async video on X, YouTube, or wherever your audience already is. That's a $50-a-month experiment, not a $50,000 one, and it tests the actual question that matters: whether people want to watch an AI-narrated recap of your niche, not whether you can keep a live feed alive.

Here's the honest counter-take. Fifty thousand viewers and 824,000 impressions for a launch-day novelty from a well-funded, well-covered company is a modest result, not proof that people want to watch AI news anchors. Impressions on a launch post mostly measure curiosity and press pickup, not sustained attention, and Misra's own framing, that the goal wasn't to be "the news" or to fool anyone, reads like someone who knows this was a capability demo dressed up as a product launch. I could be wrong if Mirage has retention data showing people actually stuck around rather than glancing and scrolling past, but nothing public so far shows that. Treat this as a useful checkpoint on synthetic media capability and cost, not as evidence that AI news anchors are a business yet.

Author

Lukas

@lukcombinator

Sources

• Mirage AI Made a 24-Hour AI News Channel. Who Is This For?
• The World's First AI News Network Just Went Live
• Mirage raises $75M to continue building models for its AI video-editing app Captions
• HeyGen Pricing in 2026: Plans, Credits, and Real Costs Explained
• The Complete Guide to ElevenLabs Plans, Overages and Usage-Based Pricing in 2026