How to tell if a robot video is real: CGI, sped up, or teleoperated?

The short answer: most robot videos that fool people aren't fake. The robot is real. What's hidden is how it was filmed. Before you believe a clip, ask four questions: is it computer graphics, is it sped up, is a person controlling it, and is this the one good take? The fastest checks are the caption, the original upload, and the company's own blog post. Companies that are proud of autonomy tend to say "autonomous" and "1x speed" out loud. When they don't say it, that's worth noticing.

Part 1: Why I keep asking "real or CGI?"

On the Servospace Shorts I ask the same thing every day: real or CGI? I started doing it half as a game. It turned into the most useful habit I've picked up since I started learning robotics.

Here's why. In my first guide I said I'd be skeptical of demos: sped up, remote-controlled, or the one good take out of fifty. I wrote that line before I'd really checked how often each of those things happens. So I went and checked. This guide is what I found.

The surprising part for me: the out-and-out fakes, the pure CGI ones, are the rare case. The common case is a real robot, filmed honestly in some narrow sense, that leaves you believing something that isn't true. A human is steering it. The footage runs at four times speed. Or the robot nailed the move once after failing it a lot.

None of that is automatically bad. Teleoperation is how a lot of robots learn. Speeding up a slow research video saves you minutes of watching. Failed takes are part of every hard thing. The problem is only when you can't tell which one you're looking at.

I'm not an engineer, and I'm not a visual-effects person either. Everything below is a checklist I can actually use on my phone, plus the real cases that taught me each item. Where I couldn't find a source, I left it out.

Part 2: The checklist

The four questions

Now the details, with the cases that made each one click for me.

1. Is it CGI?

The most famous fake robot video I know of is Corridor Digital's "Bosstown Dynamics." In June 2019, Corridor, a visual-effects studio in Los Angeles, posted a video of a humanoid robot being run through agility drills, hit with a hockey stick and bullwhipped until it turns a gun on the people testing it. It looked a lot like the real Boston Dynamics test videos it was spoofing (Gizmodo).

There was no robot. An actor performed the moves in a motion-capture suit, on green screen, and the robot was added in afterward (Xsens).

What I like about this case is how many clues were there the whole time:

• The watermark. Corridor put "Bosstown Dynamics" right in the lower-right corner of the frame (Gizmodo). Not Boston. Bosstown.
• The humans held back. Gizmodo pointed out that in the behind-the-scenes footage, the guy with the hockey stick is pulling his swings so he doesn't hurt the actor. Real Boston Dynamics testers had no reason to go easy on a machine (Gizmodo).
• The creators said so. Corridor posted a "how we made it" video showing the motion capture (Gizmodo). When it made a sequel that October, with an explainer on its Corridor Crew channel a day later, the sequel fooled people too, and Snopes had to fact-check it (Snopes).

So why did it fool so many people? Gizmodo's answer: the clip got cut down and reshared without context (Gizmodo). The reupload is where the label falls off.

That's my number one CGI check, and it's boring: find the original upload. Who posted it first? Is it a robotics company or lab, or a channel that makes visual effects? A 15-second clip on a random account is hard to judge. The same clip on its original channel, next to a "how we made it" video, is easy.

My other CGI checks are softer, and I hold them loosely:

• Does the robot interact with real things in a way that's hard to fake? Picking up a floppy object, getting shoved, kicking up dust.
• Do people in the frame react to it like it's a heavy, dangerous machine?
• Is there a version of the same moment filmed from a second angle, or live in front of a crowd?

I'm deliberately not giving you a list of "AI video tells" like weird hands or melting backgrounds. AI video tools are changing too fast, and I haven't found a source I trust that stays current. The original-upload check works no matter what tool made the clip.

2. Is it sped up?

This is the easiest one to check and the one I used to miss most.

Research labs often speed up their videos because real robots can be slow and nobody wants to watch a ten-minute pick-and-place. The honest ones label it. Princeton's TidyBot project, a robot that learns how you like your room tidied, says plainly on its project page: "All videos in this section play at 4x speed" (TidyBot). That's fine. It's disclosed. But clip the video for social media, drop the caption, and the robot suddenly looks four times more capable than it is.

What to look for:

• A label in the corner of the frame. "2x", "4x", "8x". It's often small and easy to scroll past.
• A line in the caption or project page. Search the page for "speed" or "x".
• The people and things around the robot. If a person walks through the shot, or a curtain moves, or a clock is visible, does that look normal? If the humans look jittery and fast, the robot is too.
• The opposite label. Some companies now go out of their way to say "1x speed." More on that below.

If there's no label and no people in frame to compare against, I treat the speed as unknown, not as real time.

3. Is someone driving it?

This is the big one, and it's where most of the honest confusion lives. Teleoperation means a person controls the robot remotely, often with a VR headset, gloves, or a second robot arm that the robot copies. The robot is real. The decisions belong to a human.

Case: Tesla Optimus folding a shirt (January 2024). Elon Musk posted a video of Optimus pulling a shirt from a basket and folding it. Viewers spotted a hand in a control glove at the edge of the frame, moving in sync with the robot. Musk then posted: "Important note: Optimus cannot yet do this autonomously" (New Atlas). The tell was right there in the corner of the shot.

Case: Tesla's "We, Robot" event (October 2024). At Tesla's robotaxi event near Los Angeles, Optimus robots served drinks, talked to guests and played games. Bloomberg reported, citing people familiar with the matter, that employees stationed elsewhere oversaw many of those interactions, though the robots could walk without external control (Bloomberg). At least one video from the night showed an Optimus bartender saying it was "assisted by a human" (TechCrunch). TechCrunch also noted the robots all had different voices, and their responses and hand gestures were immediate and synchronized, which is part of how people guessed. Afterward, Optimus engineer Milan Kovac wrote on X that "of course they were human assisted to some extent" (Axios).

Case: Stanford's Mobile ALOHA (January 2024). This is the one I point to as the right way to do it. The video of a wheeled two-armed robot cooking shrimp, cleaning and doing chores went viral. Right in the YouTube description, the team wrote: "The robot is teleoperated in the video. For autonomous robot skills, checkout our project website" (YouTube). Teleoperation was the point of the project. People drive the robot to collect demonstrations, and with 50 demonstrations per task the robot learned to do things like sauté and serve shrimp on its own (Mobile ALOHA). The team posted a separate compilation of the autonomous skills (YouTube). Stanford's own write-up says it directly: some of the tasks on video weren't autonomous (Stanford Report).

Same robot, two videos, clearly labeled. That's all I want.

Case: 1X NEO. 1X takes orders for NEO as a home robot and is upfront that a human may be involved. On its product page, 1X says NEO works autonomously by default, and that for tasks it doesn't know yet, an "Expert" from 1X can remotely supervise it at times the owner schedules (1X). That's useful to know before you watch any NEO clip: the honest question isn't "is a human ever involved?" It's "was a human involved in this clip?"

What to look for:

• The edges of the frame. Hands, gloves, a person in a headset, a cable. The Optimus shirt video was caught this way.
• Too-human behavior. At the Tesla event, different voices on different robots and instant replies with perfectly timed gestures are what made people suspicious.
• The caption. Look for the words "teleoperated", "autonomous", "human-assisted", or nothing at all.
• Later reporting. For big public demos, wait a few days. The Tesla details came out within a week.

4. Is this the one good take?

Even with no CGI, no speed-up and no human driver, a video can still mislead you by only showing the success.

Boston Dynamics is unusually open about this. In its 2021 post about Atlas doing parkour, the company wrote: "During filming, Atlas gets the vault right about half of the time." On the other runs, it said, Atlas makes it over the barrier but loses its balance and falls backward (Boston Dynamics). The company also put out behind-the-scenes footage of Atlas falling (TechCrunch). When the old hydraulic Atlas retired in 2024, its farewell video was partly a blooper reel (TechCrunch).

Half the time. On a move you saw land perfectly. I find that more impressive, not less, because now I know what I'm looking at.

What to look for:

• Outtakes or failure reels from the same company.
• Success rates in the blog post or paper.
• Duration. One clean run is a demo. Hours of continuous work, ideally shown live or reported by someone outside the company, is closer to a product.

What "autonomous, 1x speed" captions tell you

A lot of robot companies, especially the humanoid makers, have started labeling their own videos, and I'd learn to read the labels.

For example, when Figure posted two of its F.03 humanoids tidying a bedroom and making a bed together in May 2026, the company called it "fully autonomous," and its director of AI, Corey Lynch, stressed that the video was at 1x speed with no teleoperation (Business Insider).

Here's how I read a caption like that:

• It's a claim, not proof. It's the company describing its own video. But it's a specific, checkable claim, and a company that says it in writing is taking a risk if it's not true.
• It covers what's in the clip, not everything. "Autonomous" in a two-minute video doesn't tell you how many tries it took, or whether the room was set up just so.
• The absence is the signal. If a company usually labels its videos "autonomous, 1x" and one clip doesn't say that, I notice.

The phone checklist

When a robot clip stops my scroll, this is what I actually run through:

1. Who posted it first? Find the original company, lab or creator. Reshares lose labels.
2. Is there a speed label? Check the corners of the frame and the caption for "2x", "4x" or "1x".
3. Does it say autonomous or teleoperated? No word either way means unknown.
4. Check the edges of the frame. Hands, gloves, headsets, people standing just out of shot.
5. Do the humans in the shot move normally? They're your speed reference.
6. Is there a blog post or paper? That's where success rates and caveats live.
7. Is there a "how we made it" video? For CGI, this ends the debate.
8. Is it one take or hours of work? One clean take tells you it's possible, not that it's reliable.

Honest limits

Sometimes you can't tell, and that's fine.

A short clip with no caption, no original upload and no people in the shot can be impossible to judge from the outside. Teleoperation in particular can be invisible. The operator may be in another room, and a good one can make a robot look smooth and decisive.

When I can't tell, I try to say "unknown" instead of guessing. "Real or CGI?" is a good question. "I don't know yet, here's what I checked" is a perfectly good answer.

FAQ

Are most viral robot videos fake? Not in the CGI sense, from what I've seen. Pure fakes like Bosstown Dynamics exist, but the more common issue is a real robot shown sped up, remotely controlled, or on its best take, without a clear label.

Is teleoperation cheating? No. It's a common way to teach robots. Stanford's Mobile ALOHA team used it to collect demonstrations so the robot could learn tasks it later did autonomously. It only becomes a problem when a video lets you believe the robot is deciding on its own when it isn't.

How can you tell if a robot is remote-controlled? Check the caption for "teleoperated" or "autonomous", look at the edges of the frame for hands, gloves or headsets, and wait for later reporting on big live demos. Tesla's shirt-folding video was caught by a glove in the corner of the shot. If none of that settles it, the honest answer is "unknown."

What does "1x speed" mean on a robot video? That the footage plays in real time, not sped up. It's the company's own claim about its own video, so treat it as useful information, not a guarantee.

How do I check if a robot video is CGI? Find the original upload. Check who posted it, whether they make visual effects, and whether there's a "behind the scenes" video. Corridor Digital's Bosstown Dynamics video even had its parody name in the corner.

What I'm still figuring out

I don't have a good answer yet for the in-between cases: a robot that's mostly autonomous but gets a human nudge when it's stuck, or a task that worked on the third try in a room set up for it. Is that "real"? I think the honest answer is "real, with a footnote," and the footnote is the whole story.

If you've caught a robot video I got wrong, or you've got a tell I missed, send it to me (how to reach me is on the About page). I'll check it and add it here with credit.

Last updated November 2, 2026.

— Nate